JP2000312343A - High speed signal searching method and device, and recording medium therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a high speed signal searching method by which the same detection result as the case that an attentional window is minutely moved over entire areas of an input signal can be obtained even when the attentional window is not minutely moved over the entire areas, and to provide its device and its recording medium. SOLUTION: A characteristics template generating device 101 generates a feature quantity of a reference acoustic signal, and an acoustic signal detector 102 sets an attentional window to a received acoustic signal and detects a presence position of the acoustic signal similar to the reference acoustic signal. In this detection processing, the acoustic signal detector 102 repeats a process that generates a feature quantity series of the input acoustic signal in the attentional window, a process that calculates a similarity value between the feature quantity series of the reference acoustic signal and the generated feature quantity series from the input acoustic signal, a process that calculates a movement available quantity of the target window on the basis of the similarity value, and a process that sets the attentional window on the basis of the movement available quantity. Thus, the similarity value of the attentional window at each position is calculated and compared with a threshold value to discriminate the presence of a reference signal.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-speed signal search method and apparatus for searching a signal sequence for a location of a signal similar to a previously registered signal, and a method for recording the same.
It relates to a recording medium. The recording medium records a program for causing a computer to execute the high-speed signal search method, and is readable by a computer. The present invention can be used, for example, for sound signal detection. That is, the present invention detects the time at which a specific commercial is broadcast from the audio signal of the broadcast and automatically records the time, and detects or starts a video recording by detecting a specific theme song. Related to enabling signal detection technology. In addition, the present invention also relates to a technology that makes it possible to automatically monitor the time when a clapping sound is emitted, the time when a laughter is emitted, or search for a specific scene in broadcasting. Related. further,
The present invention is applicable not only to detection of audio signals but also to detection of general signals (such as video signals).

[0002]

2. Description of the Related Art Conventionally, there is a matched filter as a technique for detecting, from a signal sequence, a region where a target signal (hereinafter referred to as a "reference signal") exists. The matched filter calculates a correlation value between a waveform in a region of interest (hereinafter, referred to as an “interest window”) of the input signal and a waveform of the reference signal while moving the window of interest, and calculates the correlation value. When a certain value or more is exceeded, it is a method of determining that “the reference signal exists within the window of interest”.

[0003]

However, in this method, it is necessary to calculate the correlation value while finely moving the position of the window of interest with respect to the entire region of the input signal. The calculation speed becomes slow. "

On the other hand, instead of using the correlation value between the input signal waveform and the reference signal waveform, a characteristic amount of the input signal waveform in the window of interest is calculated, and the characteristic amount and the previously calculated reference signal waveform are calculated. A method of comparing an input signal waveform with a reference signal waveform using a correlation value with a feature amount, a Euclidean distance, or the like is also conceivable. However, in this method, similarly, it is necessary to calculate the feature value while moving the position of the window of interest finely with respect to the entire region of the input signal, and to compare the feature value. , The amount of calculation becomes enormous and the calculation speed becomes slow. "

SUMMARY OF THE INVENTION The present invention has been made to solve such a problem, and an object of the present invention is to provide a method of finely controlling the entire area of an input signal without moving a window of interest finely. It is to provide a high-speed signal search method and apparatus capable of obtaining a detection result equivalent to the case of moving, and a recording medium thereof. " That is, the present application proposes a technique that can increase the speed without reducing the accuracy.

Another object of the present invention is to provide a high-speed signal detecting apparatus capable of detecting a signal with higher accuracy even when the signal fluctuates due to noise or the like, or when it is difficult to distinguish the signal with average characteristics. Providing a signal search method and apparatus and a recording medium thereof ”. Further, another object of the present invention is to provide
The object of the present invention is to provide a high-speed signal search method and apparatus capable of detecting a signal with less processing than before, even when a signal is detected based on a large number of reference signals, and a recording medium therefor.

[0007]

According to the first aspect of the present invention,
A first step of generating a feature amount sequence for a pre-registered reference signal, a second step of setting an input signal attention window for the input signal, and an input signal within the input signal attention window And a third step of calculating an input signal similarity value indicating a degree of similarity between the feature amount sequence generated in the first step and the feature amount sequence generated in the third step. Step 5, a fifth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the fourth step, and a fifth step Determining the position of the input signal attention window based on the movable amount, and setting the input signal attention window at that position. By repeating, for each position of the input signal attention window, the input Calculating a signal similarity value and determining whether or not a reference signal is present at a position currently indicated by the input signal attention window on the input signal based on a comparison result between the input signal similarity value and a preset threshold value. It is characterized by.

According to a second aspect of the present invention, a first step of generating a feature amount sequence for a reference signal registered in advance, and a reference signal attention window for the feature amount sequence generated in the first step are set. A second step of setting, a third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and generating a feature amount sequence for the input signal input A fourth step, a fifth step of setting an input signal attention window for the feature amount series generated in the fourth step, and setting the input signal attention window to the plurality of reference signal division attention windows. A sixth process of dividing into a plurality of corresponding input signal division windows of interest, a feature value sequence in each reference signal division window of interest, and a feature value sequence in an input signal division window of interest corresponding to the reference signal division window of interest. The seventh step of calculating the input signal similarity value indicating the similarity degree of An eighth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the seventh step, and a movable amount calculated in the eighth step Ninth step of determining the position of the input signal attention window based on the quantity and setting the input signal attention window at that position, and repeating the ninth step from the sixth step. The input signal similarity value is calculated for each position of the input signal attention window, and the reference signal is present at the position indicated by the input signal attention window based on the comparison result between the input signal similarity value and a preset threshold value. Is determined.

According to a third aspect of the present invention, there is provided a first step of generating a feature amount sequence for a plurality of reference signals registered in advance, a second step of generating a feature amount sequence for an input signal input, A third step of setting an input signal attention window with respect to the feature amount sequence generated in the second step; and a feature amount sequence relating to one of the plurality of reference signals. A fourth step of calculating an inter-reference signal similarity value indicating a degree of similarity between the reference signal and the other reference signal, and each of the plurality of reference signals is generated in the first step. A fifth step of calculating an input signal similarity value indicating a degree of similarity between the feature amount series and the feature amount sequence in the input signal attention window; and a fifth step of calculating the similarity value between the reference signals calculated in the fourth step. Input signal similarity value calculated in the process And a position of the input signal attention window is determined on the basis of the movable amount calculated in the sixth step. And a seventh step of setting the input signal attention window at that position, and repeating the fifth to seventh steps to obtain an input signal similarity value for each position of the input signal attention window. Is calculated, based on a comparison result between the input signal similarity value and a preset threshold value,
It is characterized in that it is determined whether or not a reference signal exists at a position currently indicated by the input signal attention window.

According to a fourth aspect of the present invention, a first step of generating a feature sequence for a plurality of reference signals registered in advance, and a reference signal for each feature sequence generated in the first process. A second step of setting an attention window, a third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and a feature amount sequence for the input signal input. A fourth step of generating an input signal attention window for the feature amount sequence generated in the fourth step, and dividing the input signal attention window by the plurality of reference signal divisions. A sixth step of dividing the input signal into windows of interest corresponding to the window of interest;
Each of the reference signal divisions is a similarity value indicating a degree of similarity between a feature value sequence related to one reference signal and a feature value sequence related to the other reference signal, and corresponds to each other between the two reference signals. A seventh process of calculating a similarity value between reference signals, which is a similarity value indicating a degree of similarity between feature amount sequences in the window of interest, and, for each reference signal of the plurality of reference signals, a reference signal division in each window of interest. An input signal similarity value indicating the degree of similarity between the feature amount sequence of the above and the feature amount sequence in the input signal division attention window corresponding to the reference signal division attention window is calculated in an eighth process and a seventh process. A ninth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved, based on the similarity value between the reference signals and the input signal similarity value calculated in the eighth step; Movable calculated by Based on, to determine the position of the input signal of interest window, first setting the input signal of interest window in that position
0, and by repeating the sixth to the tenth steps, an input signal similarity value is calculated for each position of the input signal attention window, and the input signal similarity value is preset as the input signal similarity value. It is characterized in that it is determined whether or not a reference signal exists at a position currently indicated by an input signal attention window based on a result of comparison with a threshold value.

According to a fifth aspect of the present invention, in the high-speed signal search method according to any one of the first to fourth aspects, a similar value between the input signal and the reference signal is determined for a plurality of portions of the input signal. A preliminary matching process that is determined in advance, and a plurality of similar values obtained in the preliminary matching process, a preliminary matching similar value statistical process that determines an average value and a standard deviation value of the similar values;
A threshold value determining step of determining the threshold value based on the average value and the standard deviation value obtained in the preliminary matching similarity value statistical step.

According to a sixth aspect of the present invention, in the high-speed signal searching method according to the first or second aspect, the input signal similarity value is an overlap similarity value.

According to a seventh aspect of the present invention, in the high-speed signal search method according to the first or third aspect, the input signal attention window is an attention window having the same time length as the reference signal in a time axis direction. Features.

According to an eighth aspect of the present invention, in the high-speed signal search method according to any one of the first to fourth aspects, the movable amount is such that θ is the threshold value and S is the input signal similarity value. And D is the total frequency of the feature amount series, and
When or ｛｝ represents rounding down to the decimal point, it is characterized by being calculated by the equation (1).

According to a ninth aspect of the present invention, in the high-speed signal search method according to any one of the first to fourth aspects, the reference signal and the input signal are acoustic signals.

The invention according to claim 10 is the invention according to claim 2 or 4.
In the high-speed signal search method described above, the input signal attention window and the reference signal attention window are attention windows having the same time length as the reference signal in a time axis direction.

The invention according to claim 11 is the invention according to claim 3 or 4.
In the high-speed signal search method described above, the similarity value between the reference signals and the similarity value of the input signal are overlapped similarity values.

According to a twelfth aspect of the present invention, a first step of generating a feature amount sequence for a reference signal registered in advance,
A second step of creating a histogram for the feature amount sequence generated in the first step, a third step of setting an input signal attention window for an input signal input, A fourth step of generating a feature amount sequence for the input signal of step (a), a fifth step of creating a histogram for the feature amount sequence generated in the fourth step, and a histogram and a fifth step of forming a histogram of the second step. A sixth step of calculating an input signal similarity value indicating the degree of similarity with the histogram created in the step, and an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the sixth step And the position of the input signal attention window is determined based on the movement amount calculated in the seventh process, and the input signal attention window is set at that position. An eighth process, By repeating the fourth to eighth steps, an input signal similarity value is calculated for each position of the input signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold value. It is characterized in that it is determined whether or not a reference signal exists at a position currently indicated by the input signal attention window on the input signal.

According to a thirteenth aspect of the present invention, a first step of generating a feature amount sequence for a reference signal registered in advance,
A second step of setting a reference signal attention window for the feature amount sequence generated in the first step, and a reference signal attention window set in the second step as a plurality of reference signal division attention windows. A third step of dividing, a fourth step of generating a feature value sequence for the input signal input, and a fifth step of setting an input signal attention window for the feature value sequence generated in the fourth process. And a sixth step of dividing the input signal attention window into a plurality of input signal division attention windows corresponding to the plurality of reference signal division attention windows, and a feature amount sequence in each reference signal division attention window. A seventh step of creating a histogram, an eighth step of creating a histogram for a feature amount sequence in each input signal division window of interest, and a histogram in each reference signal division window of interest and corresponding to the reference signal division window of interest. In the input signal division window of interest A ninth step of calculating an input signal similarity value indicating the degree of similarity with the togram, and a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the ninth step. Based on a tenth step of calculating and the movable amount calculated in the tenth step,
An eleventh step of determining the position of the input signal attention window and setting the input signal attention window at the position, and repeating the sixth to eleventh steps to obtain the input signal attention window. For each of the positions, the input signal similarity value is calculated, and based on the comparison result between the input signal similarity value and a preset threshold value, it is determined whether or not the reference signal is present at the position currently indicated by the input signal attention window. It is characterized by doing.

According to a fourteenth aspect of the present invention, a first step of generating a feature amount sequence for a plurality of reference signals registered in advance, and a second step of generating a histogram for the feature amount sequence generated in the first step. , A third step of generating a feature amount sequence for the input signal that has been input, a fourth step of setting an input signal attention window for the feature amount sequence generated in the third step, A fifth step of creating a histogram for the feature amount series in the input signal attention window, and, for two reference signals of the plurality of reference signals, a histogram of one reference signal and a histogram of the other reference signal. A sixth step of calculating a similarity value between reference signals indicating a degree of similarity, and a histogram generated in the second step and a fifth step for each of the plurality of reference signals. 7th step of calculating an input signal similarity value indicating the degree of similarity with the histogram generated in the step, and the similarity between the reference signals calculated in the sixth step and the input signal similarity calculated in the seventh step. An eighth step of calculating a movable amount indicating the amount by which the input signal attention window can be moved based on the value and the position of the input signal attention window based on the movable amount calculated in the eighth step. A ninth step of determining and setting the input signal attention window at that position, and repeating the ninth step from the fifth step,
An input signal similarity value is calculated for each position of the input signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold, whether a reference signal is present at the position currently indicated by the input signal attention window. It is characterized in that it is determined whether or not it is.

According to a fifteenth aspect of the present invention, a first step of generating a feature value sequence for a plurality of reference signals registered in advance, and a feature value sequence generated in the first process,
A second step of setting a reference signal attention window, a third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and a feature of the input signal. A fourth step of generating a quantity sequence; a fifth step of setting an input signal attention window for the feature quantity sequence generated in the fourth step; A sixth process of dividing into a plurality of input signal division windows of interest corresponding to the signal division window of interest, a seventh process of creating a histogram for a feature amount sequence in each reference signal division of window of interest, An eighth process of creating a histogram for the feature amount series in the window, and for two reference signals of the plurality of reference signals, determining the similarity between the histogram for one reference signal and the histogram for the other reference signal. A to similar values, and the two
A ninth step of calculating a similarity value between reference signals, which is a similarity value indicating a similarity degree between histograms in the reference signal division attention windows corresponding to each other between one reference signal, and each reference among a plurality of reference signals A tenth step of calculating, for the signal, an input signal similarity value indicating a degree of similarity between the histogram in each reference signal division window of interest and the histogram in the input signal division window of interest corresponding to the reference signal division window of interest;
Calculating a movable amount indicating an amount by which the input signal attention window can be moved, based on the reference signal similarity value calculated in the ninth process and the input signal similarity value calculated in the tenth process;
Step 1 and a twelfth step of determining the position of the input signal attention window based on the movable amount calculated in the eleventh step, and setting the input signal attention window at the position. By repeating the sixth to twelfth steps, an input signal similarity value is calculated for each position of the input signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold value. It is characterized in that it is determined whether or not a reference signal is present at the position currently indicated by the input signal attention window.

The invention of claim 16 is the invention of claims 12 to 1.
5. The high-speed signal search method according to claim 5, wherein the feature amount sequence is a sequence of a number of zero crossings of a reference signal and an input signal and a derivative thereof, and the histogram is a sequence of the number of zero crossings and a derivative thereof. Is divided into a plurality of sections, and a feature amount series corresponding to each section is created by counting based on the number of zero crossings and its differential value.

The invention according to claim 17 is the invention according to claims 12 to 1.
6. In the high-speed signal search method according to any one of the first to fifth aspects, the feature amount sequence is a sequence of feature vectors having components of a plurality of frequency bands as elements, and the histogram can take each element of the feature vector. It is characterized by being created by dividing a value range into a plurality of sections and counting a feature amount series corresponding to each section based on the value of the element.

The invention according to claim 18 is a reference signal feature value generating means for generating a feature value sequence for a reference signal registered in advance, and an input signal feature value generating means for generating a feature value sequence for an input signal inputted. Input signal attention window setting means for setting an input signal attention window for the feature value sequence generated by the input signal feature value generation means; and a feature value sequence generated by the reference signal feature value generation means; Input signal similarity value calculating means for calculating an input signal similarity value indicating a degree of similarity with the feature amount sequence in the signal attention window; and an input signal similarity value calculated based on the input signal similarity value calculated by the input signal similarity value calculating means. A movable amount calculating means for calculating a movable amount indicating an amount by which the window can be moved, and a position of the input signal attention window is determined based on the movable amount calculated by the movable amount calculating means. The input signal attention window position determination means for setting the force signal attention window, the input signal similarity value calculation means, the movable amount calculation means, and the input signal attention window position determination means are made to repeatedly perform the respective processes. ,
An iterative means for calculating the input signal similarity value for each position of the input signal attention window, and the input signal attention window currently indicating on the input signal based on the comparison result between the input signal similarity value and a preset threshold value Determining means for determining whether or not the reference signal exists at the position.

According to a nineteenth aspect of the present invention, there is provided a high-speed signal search device according to the eighteenth aspect, a video device, and control means for controlling a recording operation of the video device based on a determination result of the high-speed signal search device. It is characterized by having.

According to a twentieth aspect of the present invention, there is provided a computer, comprising: a reference signal generation means for generating a characteristic amount sequence for a reference signal registered in advance; and an input signal characteristic generating means for generating a characteristic amount sequence for an input signal inputted. Quantity generation means, input signal attention window setting means for setting an input signal attention window for the feature quantity sequence generated by the input signal feature quantity generation means, and feature quantity sequence generated by the reference signal feature quantity generation means And input signal similarity calculation means for calculating an input signal similarity value indicating the degree of similarity with the feature amount sequence in the input signal attention window, based on the input signal similarity value calculated by the input signal similarity value calculation means, A movable amount calculating means for calculating a movable amount indicating an amount capable of moving the input signal attention window, and a position of the input signal attention window is determined based on the movable amount calculated by the movable amount calculating means. The input signal attention window position determination means for setting the input signal attention window at the position, the input signal similarity value calculation means, the movable amount calculation means, and the input signal attention window position determination means perform the respective processing. By repeatedly performing, for each position of the input signal attention window,
An iterative means for calculating an input signal similarity value, and based on a result of comparison between the input signal similarity value and a preset threshold, whether or not a reference signal is present on the input signal at a position currently indicated by the input signal attention window. This is a recording medium on which a program for causing the program to function as a determination unit for determining whether or not the recording medium has been recorded.

[0027]

Embodiments of the present invention will be described below with reference to the drawings. §1. First Embodiment First, a first embodiment of the present invention will be described with reference to the drawings. In the present invention, various processing target signals can be used, but here, as an example of the processing target signal,
Use acoustic signals. Further, in the present invention, various feature amounts and various similarity scales can be used. Here, as an example of the feature amount, a histogram feature of the number of zero crossings (which is considered to be relatively effective) is used. As an example of the similarity measure, an overlap similarity value of a normalized histogram is used. Here, the feature amount means one numerical value or a set of two or more numerical values (for example, a vector) representing a part or all of the characteristics of the signal to be processed.

FIG. 1 is a block diagram showing a configuration example of a high-speed signal search device to which a high-speed signal search method according to a first embodiment of the present invention is applied. FIG. 2 is a flowchart showing an operation example of the device. This device is roughly divided into
Feature template creation device 101 and acoustic signal detection device 1
02.

Here, the high-speed signal search device shown in FIG. 1 is specifically composed of a computer device comprising a CPU (central processing unit) and its peripheral circuits. The computer device functions as each device shown in FIG. 1 by being controlled by a control program recorded on a predetermined recording medium (magnetic disk, semiconductor memory, or the like).
The control program can be distributed via a communication line.

The feature template creating apparatus 101 is an apparatus that creates a dictionary template from a reference audio signal (that is, generates a feature amount sequence of a reference audio signal). The dictionary template is used when detecting an input audio signal.
Hereinafter, each unit of the feature template creating apparatus 101 will be described. The sound signal input device 103 fetches the reference sound signal as learning data and supplies the fetched reference sound signal to the feature extraction device 104. For example, the feature extraction device 104
If the subsequent processing is performed digitally, the acoustic signal input device 103 includes an A / D converter and the like. The feature extraction device 104 calculates a feature amount (the number of zero crossings) related to the audio signal based on the audio signal supplied from the audio signal input device 103 (step S101), and supplies the feature amount to the histogram creation device 105. I do.

FIG. 3 is a block diagram showing a configuration example of the feature extraction device 104. In this figure, a frame divider 113 divides an audio signal supplied from the audio signal input device 103 into fixed time intervals (for example, 11 msec), and divides the divided audio signal into a zero-crossing counter 114 and a differentiator 115.
And supply. Here, this fixed time is called “frame”
Call. The zero-crossing counter 114 counts how many times the supplied acoustic signal (waveform) crosses the zero level within the above-mentioned fixed time, and outputs the counted result to the output y1 of the feature extraction device 104.
Output as Here, “crossing the zero level” means “when the signal level changes from a positive value to a non-positive value” and “when the signal level changes from a non-positive value to a positive value”.

The differentiator 115 differentiates the acoustic signal supplied from the frame divider 113 and supplies the result of the differentiation to a zero-crossing counter 116 and a differentiator 117. Here, the differentiation is performed, for example, by taking a difference between adjacent sample values. Zero-crossing counter 11
Numeral 6 denotes a configuration equivalent to the zero-crossing counter 114, which counts how many times the signal supplied from the differentiator 115 crosses the zero level within the above-mentioned fixed time, and outputs the count result to the feature extracting device 104. Is output as output y2. The differentiator 117 is
It has a configuration equivalent to the differentiator 115, differentiates the signal supplied from the differentiator 115, and compares the result of the differentiation with the zero-crossing number counter 1
18. The zero-crossing number counter 118 has a configuration equivalent to the zero-crossing number counter 114, counts how many times the signal supplied from the differentiator 117 crosses the zero level within the above-mentioned fixed time, and calculates the counting result. Of the feature extraction device 104
Output as y3. The configuration of the feature extraction device 104 has been described above, but the configuration of the feature extraction device 104 is not limited to such a configuration.
Any configuration may be used as long as it has a function of extracting characteristics of the audio signal from the audio signal supplied to the audio signal 04. A feature can be represented by a set of several feature values such as y1, y2, and y3. In the present application, such a set of some feature amounts is referred to as a “feature vector”.

The histogram creation device 105 is a device for creating a histogram of a feature vector from a series of feature vectors for each frame supplied from the feature extraction device 104 (step S102). A histogram of feature vectors can be created by classifying feature vectors into several patterns and counting the number of feature vectors belonging to each classification. There are many possible ways to classify feature vectors into patterns. The histogram creation device 105, for example, uses the feature vector (ie,
The range of values (from the minimum value to the maximum value) that can be taken by the number of zero crossings y1, y2, y3 in each frame is divided into multiple sections, and the number of zero crossings y1, y2, y3 in each frame is The frequency of each section is counted by categorizing it into any of the sections. Therefore, in the case of this example, a histogram g having the number of sections given by the product of the number of sections of each feature amount is created. in this way,
The learning phase ends by creating a histogram for the reference audio signal. The histogram is supplied to the similarity value calculation device 110 as a template of the reference sound signal.

The acoustic signal detecting device 102 detects a position where an acoustic signal similar to the reference acoustic signal exists from the input acoustic signal sequence. Hereinafter, the acoustic signal detection device 102 will be specifically described.

The sound signal input device 106 is a device for taking in an input sound signal, and has a configuration equivalent to the sound signal input device 103. The acoustic signal captured by the acoustic signal input device 106 is supplied to the attention area scanning / extracting device 107. The attention area scanning / extracting device 107 extracts an audio signal in a partial attention area indicated by the input signal attention window from the audio signal supplied from the audio signal input device 106 using the input signal attention window (step S103), supplying the extracted sound signal to the feature extraction device 108. The attention area scanning / extracting device 107 outputs the input signal attention window (ie,
The acoustic signal is extracted while sequentially moving the position of the region of interest, and the amount of movement is calculated by the upper limit value calculation device 111. Note that the size of the input signal attention window is constant while the input signal attention window is sequentially moved.

The feature extraction device 108 is a feature extraction device 10
4, and an example of the configuration is as shown in FIG. The feature extraction device 108 calculates a feature amount related to the sound signal based on the sound signal extracted by the sound signal input device 107 (step S104).
The output of the feature extraction device 108 is the histogram creation device 1
09. The histogram creation device 109 has a configuration equivalent to that of the histogram creation device 105, and determines the frequency of the number of zero crossings for the feature amount (that is, the number of zero crossings y1, y2, y3 in each frame) supplied from the feature extraction device 108. By counting, the histogram h of the feature amount is calculated.
Is created (step S105), and the histogram h is supplied to the similarity value calculation device 110.

The similarity calculator 110 calculates the similarity between the normalized histogram h supplied from the histogram generator 109 and the normalized histogram g supplied from the feature template generator 101 (step S10).
6). As the similarity value, various definitions are possible. Here, as an example, an overlap similarity value represented by the following expression is used.

[0038]

(Equation 2)

Here, D is the total frequency of the histogram, L is the number of sections of the histogram, g _j is the value of the j-th section of the histogram g, and h _j is the value of the j-th section of the histogram h. Min (g _j , h _j ) is g
The smaller value of _j and _hj is shown. The calculated similarity value is supplied to the upper limit value calculation device 111 and the detection determination device 112.

The similarity measure indicating the degree of correlation between histograms is not limited to the above-mentioned overlap similarity value, but may be, for example, a value corresponding to the distance between histograms. Here, the “value corresponding to the distance between the histograms” is, for example, a value obtained by summing the absolute values of the frequency differences between the corresponding sections in the normalized histograms g and h over all the sections (Σ | g _j − h _j |).

The upper limit value calculating device 111 calculates the upper limit value of the similarity value at each point in the vicinity of the point at which the similarity value is obtained, based on the similarity value already obtained. For example, in the case of overlapping similar values of histograms, as shown in FIG. 4, the difference between the histogram of window A and the histogram of window B (set for the time series of the input audio signal) is It only depends on the number of samples included in the non-common area between the window and the window B. In other words, when the similarity value is determined based on a value obtained by accumulating the comparison results for each section obtained in the normalized histogram for each frame for all the sections, like the overlapping similarity value, the input signal attention window is moved. The change in the similarity value that occurs when the input signal attention window before and after the movement occupies the total number of samples (in this case, the number of frames in the input signal attention window). It will be limited to the ratio. By utilizing this property, the similarity value between the feature value of the input audio signal in a certain input signal attention window and the feature value of a certain reference signal can be calculated by using the (input audio signal in the input signal attention window) near the input signal attention window. By the similarity value between the signal features and the features of a reference signal,
The distance is limited to a certain upper limit or less (in the case of a distance, more than the lower limit).

As an example, consider a case of searching for a region having a similar value equal to or greater than a certain threshold. In this case, the feature histogram of the input audio signal at a certain time and the feature template creating apparatus 1
When the similarity value with the histogram created in FIG. 01 is obtained (matching is performed), the upper limit value of the similarity value in the vicinity of the input audio signal at the time point near the actual value is the similarity value with the histogram of the feature template at those time points. It can be obtained by calculation without obtaining (checking). That is, in a section in which the upper limit value of the similarity value is equal to or less than the threshold value, the operation for obtaining the similarity value is not necessary, and thus can be omitted. For example, in this case,
Histogram h for an input sound signal at a certain time
The similarity value obtained from the histogram and the histogram g created by the feature template creation device 101 is represented by S (step S106 in FIG. 2, processing 31 in FIG. 4), the threshold of the similarity value being searched is represented by θ, and the input signal Assuming that the number of frames in the window of interest is D, if S <θ, the movable amount of the window of interest of the input signal can be calculated as floor {D (θ−S)} + 1 frames (step in FIG. 2). S107, FIG.
Processing 32). Here, floor $ represents rounding down to the decimal point. This movable amount is supplied to the attention area scanning / extracting device 107. Attention area scan extraction device 107
Moves the input signal attention window based on the movable amount (step S108).

In this process, it is important to set an appropriate threshold value θ. However, since the appropriate threshold value θ differs depending on the input audio signal or the reference audio signal or depending on the search parameters, it is not advisable to set the threshold value θ to a single fixed value. Therefore, here, the threshold θ is determined as θ = m + cv based on the average m and the variance v of the similarity values when the search is performed in the time axis direction. Where m
And v are determined by sampling the feature vector of the input audio signal prior to the search and taking statistics of similarity values. C is a predetermined parameter.

The apparatus sequentially moves the window of interest of the input signal in this way, and repeats the same processing until the window of interest of the input signal exceeds the end of the input audio signal. At this time, if the similarity value supplied from the similarity value calculation device 110 exceeds the threshold θ, the detection determination device 112 determines that “the acoustic signal in the input signal attention window is equal to the reference acoustic signal”, and The position of the signal attention window (that is, the generation time of the acoustic signal) is output as a result (step S109). Through the above processing, the position (occurrence time) of the reference sound signal can be detected from the input sound signal.

Using this embodiment, an experiment was performed to detect a part similar to a previously registered reference sound signal from a 2-minute input sound signal actually recorded from a television broadcast.
The reference sound signal used as the registration data was obtained by separately recording and inputting a part of the input sound signal. As a reference acoustic signal, the length of the signal is 5.94 seconds, and 11.
The experiment was performed with 89 seconds each using 109 different acoustic signals. FIG. 5 shows the detection accuracy, and FIG. 6 shows the comparison of the number of times of comparison with the method of performing matching while shifting in a round robin manner.
In FIG. 5, "recall" and "accuracy" are calculated by the following equations. (Recall) = (Number of correctly detected points) / (Number of correct points) (Accuracy) = (Number of correctly detected points) / (Total number of detected points)

§2. Second Embodiment Next, a second embodiment of the present invention will be described with reference to the drawings. In the first embodiment, as a method of detecting a position of a signal similar to a previously registered reference signal, a method of calculating a feature amount of a reference signal and an input signal and determining a similar value between the two has been described. Further, in the first embodiment, the similarity value of the feature amount between the reference signal and the input signal is calculated in units of a window (attention region) having a certain time length, and the similarity value calculation is performed in the vicinity of the attention region. By determining the movement amount of the attention area based on the upper limit of the similarity value in the attention area,
A method for reducing the amount of calculation and increasing the search speed has been described.

However, in the method of the first embodiment,
There is a problem that the signal cannot be sufficiently detected when the signal fluctuates due to noise or the like, or when it is difficult to distinguish the signal with an average feature (such as an acoustic signal of an announcement).

Therefore, in the second embodiment, even when the signal fluctuates due to noise or the like, or when it is difficult to distinguish the signal with an average feature, a high-speed signal search method capable of detecting the signal with higher accuracy is provided. Will be described.

FIG. 7 is a block diagram showing a configuration example of a high-speed signal search device to which the high-speed signal search method according to the second embodiment of the present invention is applied. In the present invention, various signals to be processed can be used. Here, an acoustic signal is used as an example of the signal to be processed. This apparatus is composed of a reference feature amount extraction unit 201, an input feature amount extraction unit 202, a similar value calculation unit 203, and a movement amount calculation unit 204. The present device is provided with a pre-registered reference sound signal (ie,
A sample audio signal to be searched) and an input audio signal (that is, an audio signal to be searched) are input, and a portion where a similar value between the reference audio signal and the input audio signal exceeds a predetermined threshold θ is determined from the input audio signal. To detect.

Here, the high-speed signal search device shown in FIG. 7 is specifically constituted by a computer device comprising a CPU (central processing unit) and its peripheral circuits. The computer device functions as each unit shown in FIG. 7 by being controlled by a control program recorded on a predetermined recording medium (magnetic disk, semiconductor memory, or the like).
The control program can be distributed via a communication line.

Referring to FIG. 7, reference feature amount extraction means 201
Sets a window of interest (hereinafter, referred to as a “reference signal window of interest”) on a feature amount sequence generated from a reference audio signal, and sets the reference signal window of interest to a plurality of windows of interest (hereinafter, “reference signal division window of interest”). , And supplies the feature value sequence in each reference signal division window of interest to the similarity value calculation means 203. The input feature extraction unit 202 sets a window of interest (hereinafter, referred to as an “input signal window of interest”) on a feature value sequence generated from the input audio signal, and sets the input signal window of interest to a plurality of windows of interest (hereinafter, referred to as “windows of interest”). , Referred to as “input signal division window of interest”), and the feature value sequence in each input signal division window of interest is calculated by the similarity value calculating unit 20.
Supply to 3.

The similarity value calculating means 203 calculates a feature value sequence in each reference signal division window of interest based on the feature value sequences supplied from the reference feature value extracting means 201 and the input feature value extracting means 202 (the reference signal A similar value (hereinafter, referred to as a feature amount series) in each input signal divided window of interest corresponding to the divided window of interest.
(Referred to as "input signal similarity value"). Then, based on whether the calculated input signal similarity value is greater than the threshold value θ, the similarity value calculation unit 203 determines whether or not the reference sound signal exists at the position currently indicated by the input signal attention window. When the reference sound signal exists at the position currently indicated by the input signal attention window, the similarity value calculation unit 203 outputs the position (time or the like) currently indicated by the input signal attention window as a signal detection result. Further, the similarity value calculating means 203 supplies a similarity value necessary for calculating the movable amount of the input signal attention window to the moving amount calculating means 204.

Based on the similarity value supplied from the similarity value calculating means 203, the movement amount calculating means 204 calculates the upper limit value of the similarity value for the input signal attention window near the input signal attention window corresponding to the similar value. I do. Then, the moving amount calculating means 2
04 calculates the amount of movement of the window of interest of the input signal based on the upper limit value and calculates the amount of movement as the input feature amount extraction means 2
02. As a result, the input feature amount extraction means 2
02 indicates the position of the input signal attention window,
4 is moved by the movable amount supplied from 4. In this way, the present apparatus moves the input signal attention window sequentially,
Each of the above-described processes is similarly repeated for a feature amount sequence generated from the input audio signal in the input signal attention window after the movement.

Next, the operation of the present apparatus will be specifically described with reference to FIGS. Reference feature extraction means 201
First, a reference sound signal registered in advance is read.
Next, the reference feature extracting unit 201 extracts a feature from the read reference acoustic signal (step S2).
01). In the present invention, various feature values can be used. Here, a spectrum feature is used as an example of the feature value. In this case, the feature amount can be extracted using, for example, a band-pass filter.

A specific example of feature value extraction will be described below. For example, when it is desired to search for a specific acoustic signal of about 10 seconds from a broadcast signal of a television or a radio, a good result can be obtained by extracting a feature amount using the apparatus shown in FIG. That is, seven band-pass filters are set such that the center frequencies of the band-pass filters are equally spaced on a logarithmic axis, and a time window of 60 ms is set for the output waveform of each band-pass filter.
The average value of the square of the output waveform within the time window is calculated. A set of the seven average values obtained in this way is defined as a seven-dimensional feature vector. The feature vectors are sequentially calculated while shifting the time window by 12 milliseconds. In this case, one feature vector is obtained every 12 milliseconds. In this way, the reference feature amount extracting means 201 sequentially obtains a feature vector having components of each frequency band of the reference acoustic signal as elements. In addition, as the above-mentioned feature amount extraction method, the following method is also conceivable. That is, seven band-pass filters are set such that the center frequencies of the band-pass filters are equally spaced on a logarithmic axis, and the output of each band-pass filter is set to 12
Takes out every 60 milliseconds every millisecond, calculates the difference between the taken out output and the just taken out output, calculates the average of the squares of the difference, and calculates the obtained seven averages. This is a method of forming a set as a seven-dimensional feature vector.

Subsequently, the reference feature quantity extraction means 201 sets a window of interest for the time series of those feature vectors (step S202). The reference feature extraction means 201
First, one attention window (reference signal attention window) is set for the entire reference acoustic signal (see FIG. 8, reference feature extraction means 2).
01, see “note window”). FIG. 8 schematically shows that “the horizontal direction in the reference signal attention window corresponds to the time axis, and the feature vectors are sequentially obtained in the time axis direction”.

Next, the reference feature extraction means 201 divides the reference signal attention window into a plurality of reference signal division attention windows in the time axis direction (step S203, the reference feature extraction means 201 in FIG. 8). See “Split”). Note that the number of divisions can be determined as appropriate according to the form of the signal to be processed and the like, but here, for example, eight divisions. As a result, the reference feature value extraction unit 201 determines the feature value (ie, the time series of feature vectors) included in each reference signal division window of interest.
Are sequentially supplied to the similarity calculation means 203.

The input characteristic amount extracting means 202 first reads an input audio signal. Next, the input feature amount extraction means 20
2 extracts a feature amount from the read input audio signal (step S204). Here, the input feature amount extraction unit 202 extracts a feature amount (that is, a time series of seven-dimensional feature vectors) by the same method as the reference feature amount extraction unit 201 (see FIG. 10). Therefore, the description is omitted here.

Subsequently, the input feature quantity extracting means 202 sets a window of interest for the time series of those feature vectors (step S205). The input feature amount extraction means 202
First, an attention window (input signal attention window) having the same length as the reference signal attention window is set for the input audio signal (in FIG. 8,
(See "attention window" in the input feature amount extraction means 202). still,
FIG. 8 schematically illustrates that “the horizontal direction in the input signal attention window corresponds to the time axis, and the feature vectors are sequentially obtained in the time axis direction”.

Note that the position of the input signal attention window is set at the beginning of the feature amount sequence created from the input audio signal at the beginning of the process. The upper part is sequentially moved in the time axis direction. This moving amount is calculated by the moving amount calculating means 204.

Subsequently, the input feature extraction unit 202 divides the input signal attention window into a plurality of input signal division attention windows in the time axis direction (step S206, the input feature extraction unit 202 in FIG. 8). See “Split”). Note that the number of divisions of the input signal attention window is equal to the number of divisions of the reference signal attention window. As a result, the input feature amount extraction unit 202 sequentially outputs the feature amounts (ie, time series of feature vectors) included in each input signal division window of interest to the similarity value calculation unit 203.

The similarity value calculating means 203 sequentially reads the time series of the feature vectors divided by the reference signal dividing attention window from the reference feature extracting means 201, and reads the input signal dividing attention from the input feature extracting means 202. The time series of the feature vectors divided by the window is sequentially read.

Next, the similarity value calculating means 203 creates a histogram of the feature vector based on the time series of the feature vector in the reference signal division window of interest (step S20).
7) A histogram of the feature vector is created based on the time series of the feature vector in the input signal division window of interest (step S208). These histograms are
Each of them is created by dividing the range of values (from the minimum value to the maximum value) that each element of the feature vector can take into a plurality of bins (sections). For example, if the range of values that each element can take is divided into three sections (tentatively, “sections A, B, and C”), and if the number of elements of each feature vector is 7, then 7 Each of the elements of section A, B, C
Will belong to any of the above. Therefore, when considering one feature vector, the combination of the elements of the feature vector indicates that all seven elements belong to the section A (A, A, A, A, A, A, A). 7
Elements belong to the section C (C, C,
Up to C, C, C, C, C), a total of 3 7 combinations are conceivable. From the above, the total number of bins of the histogram of the feature vector (that is, the number of sections arranged on the horizontal axis of the histogram) is 3 7. Therefore, when the horizontal axis of the histogram is set in this manner, each feature vector is classified into any one of the 3 7 sections.

According to the method described above, the similarity value calculating means 203 is supplied with the feature vector supplied from the reference feature value extracting means 201 (divided in the time axis direction) and the feature vector supplied from the input feature value extracting means 202 (time value). A histogram is created for each of the feature vectors (divided in the axial direction). Here, these histograms are represented by G ¹ ,
G ^2, ..., G ⁿ and H ^1, H ^2, ..., and H ^n. Here, n is the number of divisions of the original window of interest (the reference signal window of interest and the input signal window of interest) (hereinafter, 1 to n are referred to as “division numbers”), and G is created from the feature vector of the reference sound signal. H indicates a histogram (hereinafter, referred to as “reference signal histogram”), and H indicates a histogram created from the feature vector of the input audio signal (hereinafter, referred to as “input signal histogram”).

Subsequently, the similarity value calculating means 203 calculates the similarity value between the input signal histogram and the reference signal histogram having the same division number (step S209). In the present invention, various similar values can be used. Here, an overlapping similar value is used as an example of the similar value. Here, the similarity value S _k (between the reference signal histogram and the input signal histogram) at the division number k is defined by the following equation.

[0066]

(Equation 3)

Here, D _k is the total frequency of the histogram of the division number k, L is the number of sections of the histogram (3 to the seventh power in the above example), and g ^k _j is the k-th histogram G
^k is the value of the j-th section, h ^k _j is the value of the j-th section of the k-th histogram H ^k , and min (g ^k _j , h ^k
_j ) indicates the smaller value of g ^k _j and h ^k _j .

Also, the similarity value (ie, the entirety of the window of interest)
Similarity between reference signal attention window and input signal attention window) S
Is defined by the following equation (hereinafter, the similarity value S is referred to as “overall similarity value”).

[0069]

(Equation 4)

Here, min (S ₁ , S ₂ ,..., S _n )
_{Is, S 1, S 2, ...} , represent the minimum value of S _n.

The calculation of the similarity value S _k is performed one by one (for example,
(In ascending order of the value k). If the calculated similarity value _Sk is equal to or smaller than the threshold value θ, it is apparent that the minimum value of the overall similarity value S is equal to or smaller than the threshold value θ based on the equation (4). The calculation of _{k + 1} , _{Sk + 2} , _{Sk + 3} ... no longer needs to be performed. Then, the similarity value calculation means 203 supplies the minimum value of the similarity values calculated so far to the movement amount calculation means 204.

[0072] On the other hand, if, as long as all similarity values S ₁ to S _n is larger than the threshold value theta, overall similarity value S becomes larger than the threshold value theta. This means that “a reference audio signal has been detected at the current position of the input signal attention window for (the feature vector of) the input audio signal”. Therefore, the similarity value calculating means 203 outputs the current position (time) as a part of the signal detection result. Also in this case, the similarity value calculating means 203 converts the overall similarity value S into the moving amount calculating means 204.
To supply.

The movement amount calculating means 204 first reads the overall similarity value S supplied from the similarity value calculating means 203. Next, the movement amount calculation means 204 calculates the possible movement amount w (step S210). Here, the movable amount w
Is determined by the following equation.

[0074]

(Equation 5)

Here, the unit of the movable amount w is the number of feature vectors, floor () represents rounding down below the decimal point, and D is the k-th divided window of interest (see S = S _k ). Is the total frequency of the histogram of the feature vector in the signal division attention window and the input signal division attention window, and θ is the above threshold.

Equation 5 indicates that if S <θ at the present time, even if the input signal attention window is moved by (w−1) feature vectors, the overall similarity value S never exceeds the threshold value θ. It means that. This is because, when the input signal attention window is moved, all the feature vectors that go out of the input signal attention window do not contribute to the overlap of the histogram, and the feature vectors that enter the input signal attention window It can be easily understood by considering the case where all of the feature vectors that come are those that contribute to the overlap of the histograms (that is, the case where the overall similarity value S reaches the threshold value θ first). That is, since the similarity value increases most in such a case, assuming such a case, the input signal attention window is set to the feature vector (w−
1) The upper limit value of the overall similarity value S when moved by the number is the threshold value θ. From this, if S <θ, the feature vector w is likely to cause the overall similarity value S to exceed the threshold θ.
The movement amount for each piece is set as the movable amount.

On the other hand, if S ≧ θ, in order to find a local peak of a similar value, w = 1 is set, and the search is performed without skipping the window of interest.

The movable amount w output from the moving amount calculating means 204 is supplied to the input feature amount extracting means 202.
The input feature amount extraction means 202 moves the input signal attention window by w feature vectors (step S211; see the dashed attention window in the input feature amount extraction means 202 in FIG. 8). Thereafter, the present apparatus similarly repeats the above-described processing (setting of a plurality of divided windows of interest, creation of a histogram, calculation of similarity values, calculation of a movable amount, and the like) (step S21).
2). If the input signal attention window moves beyond the end of the input audio signal (the time series of the feature vector) due to the movement of the input signal attention window, it means that all the input audio signals have been searched, and thus the search processing ends. I do.

Next, an operation experiment example of the high-speed signal search device according to the present embodiment will be described. Connect this device to a workstation (S
FIG. 11 and FIG. 12 show the results of implementing an audio signal on a television broadcast as a subject, implemented in GI Company O ₂ ). In each case, the vertical axis indicates a similar value, and the horizontal axis indicates time. These are those in which a commercial (15 seconds) is used as a reference signal and an actual recording of a television broadcast (January 22, 1998, 18: 22: 6 hours) is detected as an input signal. In this experiment, the sampling frequency = 11.02
5 kHz, the dimension of the feature vector = 7, and the number of bins of each feature vector element = 3.

FIG. 11 shows a case where the window of interest is not divided.
That is, FIG. 12 shows a case where the present method is not executed, and FIG. In each of the drawings, a portion marked with a circle indicates a location searched by the apparatus. According to human confirmation, the correct answers in this experiment were 18:48, 23:22:00, and 00:11. In FIG. 11, three extra locations are detected in addition to the correct location, whereas in FIG. 12, correct detection results are obtained.

In general, the larger the ratio of the similar value in the portion to be searched to the similar value in the portion not to be searched, the larger the margin for setting the set value, so that a stable search is possible. In this experiment, the value of the ratio is larger in the case of FIG. 12 than in the case of FIG. 11, and it can be seen that the effect of dividing the window of interest is exhibited.

FIGS. 13 and 14 show the results of signal detection on the image pattern. In each case, the vertical axis indicates a similar value, and the horizontal axis indicates time. These are shown in FIG. 11 and FIG.
Color information obtained by recording an actual television broadcast using color information obtained from an image of a commercial (15 seconds) different from 2 (6 hours from 18:22 on January 22, 1998) Is detected as an input signal. In this experiment, the dimension of the feature vector was set to 3, and the number of bins of each feature vector element was set to 8.

FIG. 13 shows a case where the window of interest is not divided (that is, a case where the present method is not performed), and FIG. 14 shows a case where the window of interest is divided into eight by executing the present method. In each case, the portions marked with a circle indicate the locations searched by the apparatus. According to human confirmation, the correct answers in this experiment were 18:48, 19:56, 20:35, and 22:54. In FIG. 13, one (20:35) of the correct answer points has not been detected, and two extra points have been detected, whereas in FIG. 14, correct detection results have been obtained.

§3. Third Embodiment Next, a third embodiment of the present invention will be described with reference to the drawings.

In the first and second embodiments, the method of detecting the position of a signal similar to a previously registered reference signal has been described. However, these methods include:
There is a problem that "in the case of detecting signals based on a large number of reference signals, it is necessary to repeat the detection processing by the number of reference signals, which increases the amount of calculation and decreases the detection speed".

Therefore, in the third embodiment, a description will be given of a high-speed signal search method capable of detecting a signal with less processing than in the related art even when a signal is detected based on a large number of reference signals.

FIG. 15 is a block diagram showing a configuration example of a high-speed signal search device to which the high-speed signal search method according to the third embodiment of the present invention is applied. In the present invention, various signals to be processed can be used. Here, an acoustic signal is used as an example of the signal to be processed. In FIG. 15, the present apparatus 1 includes a reference feature value calculation unit 310, an input feature value calculation unit 312, a reference signal similarity value calculation unit 314, a similar value calculation unit 316, and a movable amount calculation unit 318.

Here, the high-speed signal search device shown in FIG. 15 is specifically composed of a computer device comprising a CPU (central processing unit) and its peripheral circuits. The computer device functions as each unit shown in FIG. 15 by being controlled by a control program recorded on a predetermined recording medium (magnetic disk, semiconductor memory, or the like). The control program can be distributed via a communication line.

The present apparatus 1 receives a reference audio signal (that is, an audio signal to be searched as a sample) and an input audio signal (that is, an audio signal to be searched) which are registered in advance, and inputs the reference audio signal and the input audio signal. A point where a similar value to the signal exceeds a predetermined threshold θ is detected from the input acoustic signal.

The reference feature value calculating means 310 generates a feature value sequence for a plurality of reference sound signals. In addition, the input feature value calculation unit 312 generates a feature value sequence from the input audio signal, and sets a window of interest (hereinafter, referred to as an “input signal attention window”) on the feature value sequence.

The reference signal similarity value calculating means 314 calculates a similarity value (hereinafter referred to as “reference signal similarity value”) between (the characteristic amount series of) each reference audio signal. The similarity value calculating means 316 calculates a similarity value between the feature amount series generated by the reference feature amount calculating means 310 and the feature amount sequence in the input signal attention window set by the input feature amount calculating means 312 (hereinafter, “input signal "Similar value").

The movable amount calculating means 318 calculates the input signal similarity value based on the reference signal similarity value calculated by the reference signal similarity value calculating means 314 and the input signal similarity value calculated by the similarity value calculating means 316. A parameter (for example, an upper limit of the similarity value) that defines the range of the input signal similarity value (between the input audio signal and the reference audio signal) in the input signal attentional window near the input signal attentional window corresponding to the similar value calculate. Then, the movable amount calculating means 318 calculates the movable amount of the input signal attention window based on the value of the parameter.

Next, the operation of the present apparatus will be specifically described with reference to FIG. For the sake of simplicity, in the following description, the lengths of all reference sound signals are assumed to be equal. In practice, the present invention can be applied even when the lengths of all the reference sound signals are not always equal.

First, the reference feature value calculating means 310 reads all the given reference sound signals. Next, the reference feature value calculation means 310 extracts a feature value from the read reference sound signal (step S301). In the present invention, various feature values can be used. Here, a spectrum feature is used as an example of the feature value.
In this case, the reference feature value calculation means 310 is the same as the reference feature value extraction means 201 of the second embodiment (see FIG. 10).
Then, a feature amount (that is, a time series of seven-dimensional feature vectors) is extracted. Therefore, the description is omitted here.

Next, the reference feature value calculating means 310 creates a histogram of the feature vector based on the time series of the feature vector (step S302). here,
The histogram indicates a range of values (from the minimum value to the maximum value) that each element of the feature vector can take, by a plurality of bins (sections).
Created by dividing into The reference feature value calculating means 310 creates a histogram in the same manner as the similar value calculating means 203 (second embodiment). Therefore, the description is omitted here.

In the present embodiment, a histogram is created for the entirety of one reference acoustic signal (a time series of feature vectors thereof). In addition, as shown in the second embodiment, a histogram is created. The audio signal (the time series of the feature vectors thereof) may be divided in the time direction, and a histogram may be created for each of the divided reference audio signals (the time series of the feature vectors thereof). The number of divisions may be, for example, four. In this case, four histograms are created for one reference sound signal.

First, the input feature value calculation means 312 reads an input audio signal. Next, the input feature amount calculating means 31
2 extracts a feature amount from the read input audio signal (step S303). Here, the input feature value calculating means 312 extracts a feature value in the same manner as the reference feature value calculating means 310.

Subsequently, the input feature value calculation means 312 sets an input signal attention window having the same length as the reference acoustic signal given to the reference feature value calculation means 310 to the extracted feature value (time series of feature vector). (Step S30)
4). Note that the position of the input signal attention window is initially set at the beginning of the process.
It is set at the beginning of the feature amount sequence created from the input audio signal, and is sequentially moved in the time axis direction on the feature amount sequence as the process proceeds, by a method described later. This movable amount is calculated by the similarity value calculation means 316.

Next, the input feature quantity calculation means 312 creates a histogram of the feature vector based on the time series of the feature vector in the input signal attention window (step S30).
5). Here, the histogram is created by dividing the range of possible values (from the minimum value to the maximum value) of each element of the feature vector into a plurality of bins (sections). The input feature value calculation means 312 is a reference feature value calculation means 310.
A histogram is created in the same manner as described above. Therefore, the description is omitted here.

It should be noted that the reference feature value calculating means 310
When the reference acoustic signal is divided in the time axis direction (similar to the embodiment), the input feature amount calculating unit 312 also divides the input signal attention window (the time series of the feature vectors therein) in the time axis direction. The number of divisions is the same as the number of divisions in the reference feature value calculation means 10. Therefore, as many histograms as the number of histograms generated from one reference audio signal in the reference feature value calculating unit 310 are generated from the input audio signals in the input signal attention window.

The reference signal similarity value calculating means 314 first reads the histogram (histogram of each reference sound signal) supplied from the reference feature value calculating means 310. Here, for the sake of simplicity, consider the case where the reference audio signal is not divided (however, this is to prevent the handling of subscripts from becoming complicated. It is also applicable to cases where Here, it is assumed that the histogram supplied from the reference feature value calculating means 310 is G ¹ , G ² ,..., G ^N. Here, N is the number of reference sound signals.

The inter-reference-signal similarity value calculation means 314 calculates the similarity value between the two extracted reference audio signals for all combinations of extracting two reference audio signals from the N reference audio signals (step S306). ). Here, the similarity S (g, h) between the histogram g of the k-th reference audio signal and the histogram h of the m-th reference audio signal.
Is defined by the following equation:

[0103]

(Equation 6)

Here, D is the total frequency of the histogram.
L is the number of sections of the histogram (3 in the above example is 7
) And g_jIs the value of the j-th section of the histogram g
And h_jIs the value of the j-th section of the histogram h
, Min (g_j, H_j) Is g_jAnd h _jThe smaller of
Shows the value of

It should be noted that the reference feature value calculating means 310
When the reference audio signal is divided in the time axis direction (similar to the embodiment), the inter-reference-signal similarity value calculation unit 314 performs similarity calculation on divided portions of the two reference audio signals that correspond to each other. The result of the similar value calculation is stored in storage means (not shown).
And is supplied to the movable amount calculating means 318.

The similarity value calculation means 316 first reads the histogram supplied from the reference feature value calculation means 310 and the histogram supplied from the input feature value calculation means 312. Here, for the sake of simplicity, consider a case where the reference audio signal and the input audio signal are not divided (however, this is to prevent the handling of the subscript from becoming complicated, and the present invention is applied to the case where the reference audio signal and the input audio signal are not divided). It is also applicable to a case where the signal and the input acoustic signal are divided.) Here, the histogram supplied from the reference feature amount calculating unit ^{310, G 1, G 2,} ......, and G ^N, the histogram supplied from the input feature value calculating means 312 and H. Here, N is the number of reference sound signals.

Subsequently, the similarity value calculating means 316 selects one histogram of the reference audio signal, and calculates a similarity value between the selected histogram and the histogram of the input audio signal (step S307). Similarity value S between histogram g of reference audio signal and histogram h of input audio signal
(G, h) is defined by the following equation.

[0108]

(Equation 7)

Here, D is the total frequency of the histogram.
L is the number of sections of the histogram (3 in the above example is 7
) And g_jIs the value of the j-th section of the histogram g
And h_jIs the value of the j-th section of the histogram h
, Min (g_j, H_j) Is g_jAnd h _jThe smaller of
Shows the value of

In the present embodiment, N reference sound signals are inputted, but the similarity is calculated at a time by one reference sound signal.
Done for only one. In selecting a reference sound signal used for calculating the similar value, a reference sound signal whose upper limit value of the similar value at that time may possibly exceed the threshold value θ is selected. Which reference sound signal corresponds to this can be determined by the movable amount calculating means 318 described later, and the similarity value calculating means 316 acquires the information from the movable amount calculating means 318.

When the reference sound signal and the input sound signal are divided in the time axis direction (similar to the second embodiment), the calculation of the similarity value is performed based on the (divided) reference sound signal and the input sound signal. This is done only for one segment of the sound signal.

The similarity value calculating means 316 supplies the calculated similarity value to the movable amount calculating means 318. In addition, if the similarity value exceeds the threshold θ (if the reference audio signal and the input audio signal are divided in the time axis direction, the similarity This means that the reference audio signal is present in the input audio signal,
The similarity value calculation means 316 outputs the number of the reference sound signal and the current position of the input signal attention window as the signal detection result.

The movable amount calculating means 318 first calculates the similarity value S ^k between the input audio signal and the k-th reference audio signal,
It is read from the similarity value calculation means 316. Then, the movable amount calculating means 318 calculates the movable weight w _k (step S308). Here, the movable amount w _k is obtained by the following equation.

[0114]

(Equation 8)

Here, the unit of the movable amount w _k is the number of feature vectors, floor () represents rounding down below the decimal point, and D is the k-th reference sound signal satisfying S = S ^k . The total frequency of the histogram of the feature vector, θ
Is the above threshold value.

Equation 8 indicates that if S ^k <θ at this time, even if the input signal attention window is moved by the amount of feature vectors (w _k -1), the similarity value S ^k will never exceed the threshold θ. No. " This is because, when the input signal attention window is moved, all the feature vectors that go out of the input signal attention window do not contribute to the overlap of the histogram, and the input signal attention window If all incoming feature vectors contribute to the histogram overlap (ie, the similarity S ^k is the earliest threshold θ
Is reached)), it is easily understood. That is, since the similarity value in this case is most increased, assuming a case such, the upper limit of the similarity values S ^k when moving only the feature vector (w _k -1) pieces worth an input signal of interest window The value becomes the threshold θ.

On the other hand, if S ^k ≧ θ, w _k = 1 is set and a search is performed without skipping the window of interest in order to find a local peak of a similar value.

The above-described processing is processing for one (ie, k-th) reference sound signal. When there are a large number of reference audio signals, in the conventional method, this process is simply repeated by the number of reference audio signals, thereby obtaining a movable amount for each reference audio signal (this is called an iterative method). On the other hand, the present invention is characterized in that, when a similar value S ^k relating to a k-th reference acoustic signal is obtained, a movable amount relating to a reference acoustic signal other than the k-th is also calculated based on the similar value S ^k. And Therefore, according to the present invention, it is possible to reduce the number of times of calculating the similarity value as compared with the above iterative method.

The method for reducing the number of times of calculating the similarity value is as follows.
This is described below. Two reference sound signals R^k, R^mFrom each
H histogram G^k, G^mIs created and the input sound signal I
It is assumed that a histogram H has been created. However, hiss
Togram G^k, G^mAre equal and the reference audio signal
Signal and input audio signal are not divided in the time axis direction
Shall be. Also, the histogram G ^kAnd histogram H
Input signal similarity value S^kHas already been obtained.

At this time, if the histogram G^kAnd hiss
Togram G^mAre very similar, the histogram H
And histogram G^mInput signal similarity value S^mCalculate
The input signal similarity value S^mIs obtained. Immediately
C, histogram G^kAnd histogram G^mBetween the reference signal and
The similarity value is S^kmThen the histogram G^kAnd histogra
Mu G ^mAre all input signal similarity values S^mincrease of
, The input signal similarity value S^mIs the upper limit
Reach. This is represented by the following equation:

[0121]

(Equation 9)

However, since S ^m ≦ 1 from the beginning, the expression of Expression 9 is valid only when S ^k <S ^km .

On the other hand, if the histogram G ^k and the histogram H are very similar, all the mismatched portions between the histogram G ^k and the histogram H are equal to the input signal similarity value S.
^When contributing to an increase in ^m , the input signal similarity value S ^m reaches its upper limit. This is represented by the following equation:

[0124]

(Equation 10)

However, since S ^m ≦ 1 from the beginning, the equation (10) is valid only when S ^k > S ^km . When the expressions of Expression 9 and Expression 10 are put together, the following expression is obtained.

[0126]

[Equation 11]

[0127] That is, the greater the difference between the reference signal between similar values S ^miles an input signal similarity value S ^k, the upper limit value of the input signal similarity value S ^m is smaller. At this time, the movable amount w _m
Is calculated by the following equation.

[0128]

(Equation 12)

[0129] That is, the smaller the upper limit value of the input signal similar value S ^m, movable amount w _m increases.

As described above, according to the present invention, when the reference signal similarity S ^km between two reference sound signals R ^k and R ^m is known, the reference sound signal R ^k and the input sound signal I When the input signal similarity value S ^k is obtained based on the input signal similarity values S ^k, not only the movable amount related reference acoustic signal R ^k, also be determined movable amount about the reference acoustic signal R ^m it can. Therefore, every time an input signal similarity value S ^k between the k-th reference audio signal R ^k and the input audio signal I is obtained (in the similarity value calculation means 316), the movable amount calculation means 318 outputs the reference audio signal R ^k to the reference acoustic signal a reference signal between similar values of the reference acoustic signal other than R ^k (from the reference signal between similar value calculating means 314) is taken out, the reference signal between similar values and number 11, using a number 12 expression , (The input signal attention window) are determined for all the reference sound signals. Then, when the movable amount obtained in this way is larger than the movable amount obtained before then, the movable amount calculating means 318 updates the movable amount.

The similarity value calculation means 316 supplies the updated movable amount w _i (i = 1,..., N) to the input feature value calculation means 312. The input feature value calculation means 312 moves the input signal attention window by the smallest movable amount of the movable amount w _i (step S309). Thereafter, the present apparatus similarly repeats the above-described processing (creation of a histogram, calculation of a similar value, calculation of a movable amount, and the like) (step S310). Also, by moving the input signal attention window,
If the input signal attention window exceeds the end of (the time series of the feature vector of) the input audio signal, it means that the entire input audio signal has been searched, and the search processing ends.

Next, an operation example of the high-speed signal search apparatus according to the present embodiment will be described. In order to investigate the effect of the present invention, a search experiment was performed using a 6-hour TV broadcast sound signal as an input signal and a 15-second CM sound signal as a reference signal.
Since the present apparatus is the same in accuracy as when the reference signals are separately collated, only the number of collations is compared. In addition, since the full search is always performed when the similarity value exceeds the threshold value, the number of times the similarity value is equal to or less than the threshold value is
The ratio (referred to in the present application as the number of times of collation) is compared with the case where the reference signals are collated separately. The search parameter is sampling frequency = 1
1.025 kHz, number of band filters = 7, analysis window length of frequency analysis = 60 ms, analysis window shift = 10 ms, number of bins in each feature dimension = 3, number of time window division = 1, threshold value θ
= 0.8. FIG. 17 shows an example of the search result. The vertical axis indicates similar values, and the horizontal axis indicates time. In addition, a circle indicates a detection position, and a broken line indicates a threshold value θ (0.8).

FIG. 18 shows the experimental results. In the experimental example (a) of FIG. 18, when three CMs of different products selected at random are used as reference signals, in the experimental example (b), three CMs of the same product having extremely similar sounds are used as reference signals. This is the result of the case. The similarity values between the reference signals were 0.11, 0.22, and 0.23 in the experimental example (a), and 0.72, 0.75, and 0.88 in the experimental example (b). Experimental example (a)
In the case of (1), the effect of reducing the number of times of collation is small (99.9% of the number of times of collation). (62.7% of the number of times of collation). As described above, the present apparatus is particularly effective when the similarity between the reference signals is high.

§4. Supplement Although the embodiment of the present invention has been described in detail with reference to the drawings, the specific configuration is not limited to the embodiment, and there may be a design change or the like without departing from the gist of the present invention. Even this is included in the present invention.

For example, in the first to third embodiments, the description has been given of the high-speed signal search method characterized by “similar value calculation using histogram” and “movable amount calculation based on similar value”. It is not always necessary to use both of these two features together, and even if only one of them is used, a special effect can be obtained as compared with the related art.

In the first to third embodiments, the acoustic signal is used as the processing target signal. However, as described in the experimental examples relating to FIGS. 13 and 14, the color information (R, G, B) of the image is used.
Etc.) may be used.

Further, as an example of the feature value, the zero crossing number is used in the first embodiment, and the spectrum feature is used in the second and third embodiments. However, the feature value used in the present invention is not limited to these. , Other feature values can be used.

Further, in each of the above-described embodiments, the case where signal detection is performed based on the similarity value of the histogram has been described. However, signal detection is performed using a distance (Euclidean distance, L1 distance, etc.) instead of the similarity value. May go.
In such a case, it is needless to say that the same effect can be obtained by inverting the magnitude relation of the values based on the similar value.

In each of the embodiments described above, the high-speed signal search apparatus is shown as a single unit. However, the apparatus can be used to configure an automatic video control apparatus. That is, the present high-speed signal searching device is combined with a video device, and a specific sound signal (theme song, etc.)
Alternatively, control means for detecting the occurrence of a specific image pattern or the like and controlling the recording function or the like of the video device according to the detection result is provided. As a result, it is possible to realize an automatic video control device that automatically activates the video recording function, stops the recording function, or records index information at the start and end of a specific program.

Further, this makes it possible to detect the time at which a specific commercial is broadcast out of the audio signal of the broadcast and perform automatic recording, or to detect or start a video recording by detecting a specific theme song. It is possible to do.
It is also possible to automatically monitor the time when a clapping sound is emitted from the broadcast, the time when a laughter is emitted, or search for a specific scene. Further, these processes can be applied not only to detection of an acoustic signal but also to detection of a general signal such as an image pattern.

[0141]

As described above, according to the present invention,
A window of interest is set for an input signal, a feature sequence is generated for the input signal within the window of interest, and a similarity calculation is performed between the feature sequence and a feature sequence of a reference signal registered in advance. Repeating the processing of the feature amount series generation and the similarity value calculation while efficiently changing the movement amount of the position of the attention window, and detecting the reference signal position in the input signal by comparing the calculated similarity value with a threshold value Therefore, even if the window of interest is not finely moved over the entire region of the input signal, a detection result equivalent to that obtained by finely moving the window over the entire region can be obtained at high speed. Further, according to the present invention, it is possible to search for a specific music or commercial appearance time from a long-time signal such as a broadcast, and to search for a signal including a specific signal from a signal database or the like. .

Further, according to the present invention, it is possible to perform a search for a feature amount sequence in a plurality of attention windows, taking into account the context (on the time axis) between these feature amount sequences. This makes it possible to detect a signal with higher accuracy even when the signal fluctuates due to noise or the like, or when it is difficult to distinguish the signal with average characteristics. As the setting form of the “plurality of windows of interest”, for example, if the original window of interest is divided and set as a plurality of windows of interest, the search can be performed in consideration of the context of time. become.

Further, according to the present invention, even when a signal is detected based on a large number of reference signals, the signal can be detected with less processing, and the calculation efficiency in the signal detection processing can be improved. Note that, by reading and executing the program recorded on the recording medium of the present invention by a computer system, signals can be detected with less processing, and the calculation efficiency in the signal detection processing can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration example of a high-speed signal search device according to a first embodiment of the present invention.

FIG. 2 is a flowchart illustrating an operation example of the high-speed signal search device according to the first embodiment of the present invention.

FIG. 3 is a block diagram illustrating a configuration example of a feature extraction device 104.

FIG. 4 is an explanatory diagram showing an example of similarity value calculation before and after movement of a window.

FIG. 5 is a table showing an example of detection accuracy of the high-speed signal search device according to the first embodiment of the present invention.

FIG. 6 is a table showing an example of a detection speed (speed ratio with a brute force method) of the high-speed signal search device according to the first embodiment of the present invention.

FIG. 7 is a block diagram illustrating a configuration example of a high-speed signal search device according to a second embodiment of the present invention.

FIG. 8 is an explanatory diagram illustrating a processing example of a high-speed signal search device according to a second embodiment of the present invention.

FIG. 9 is a flowchart illustrating an operation example of the high-speed signal search device according to the second embodiment of the present invention.

FIG. 10 is a block diagram illustrating a configuration example for extracting a spectral feature.

FIG. 11 is a graph showing an example of an acoustic signal detection result according to a conventional method.

FIG. 12 is a graph showing an example of an acoustic signal detection result according to the second embodiment of the present invention.

FIG. 13 is a graph showing an example of an image pattern detection result according to a conventional method.

FIG. 14 is a graph showing an example of an image pattern detection result according to the second embodiment of the present invention.

FIG. 15 is a block diagram illustrating a configuration example of a high-speed signal search device according to a third embodiment of the present invention.

FIG. 16 is a flowchart illustrating an operation example of the high-speed signal search device according to the third embodiment of the present invention.

FIG. 17 is a graph showing an example of an acoustic signal detection result according to the third embodiment of the present invention.

FIG. 18 is a graph illustrating an example of a relationship between a similar value between reference signals and an effect of reducing the number of times of matching in the third embodiment of the present invention.

[Explanation of symbols]

 101 Feature Template Creation Device 102 Acoustic Signal Detection Device 103 Acoustic Signal Input Device 104 Feature Extraction Device 105 Histogram Creation Device 106 Acoustic Signal Input Device 107 Attention Area Scanning Extraction Device 108 Feature Extraction Device 109 Histogram Creation Device 110 Similar Value Calculation Device 111 Upper Limit Calculation device 112 Detection / determination device 113 Frame divider 114 Zero-crossing number counter 115 Differentiator 116 Zero-crossing number counter 117 Differentiator 118 Zero-crossing number counter 201 Reference feature amount extraction means 202 Input feature amount extraction means 203 Similar value calculation Means 204 Movement amount calculation means 301 Signal detection device 310 Reference feature amount calculation means 312 Input feature amount calculation means 314 Reference signal similarity value calculation means 316 Similar value calculation means 318 Movable amount calculation means

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] March 13, 2000 (200.3.1)
3)

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Claims

[Correction method] Change

[Correction contents]

[Claims]

3. A first step of generating a feature sequence for a reference signal registered in advance, and a second process of setting a reference signal attention window for the feature sequence generated in the first process. A third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and a fourth step of generating a feature amount sequence for the input signal input.
And a fifth step of setting an input signal attention window for the feature amount sequence generated in the fourth step. The input signal attention window corresponds to the plurality of reference signal division attention windows. A sixth process of dividing into a plurality of input signal division windows of interest, and a similarity between a feature sequence in each reference signal division window of interest and a feature sequence in the input signal division window of interest corresponding to the reference signal division window of interest. A seventh step of calculating an input signal similarity value indicating the degree; and an eighth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the seventh step. And a ninth step of determining the position of the input signal attention window based on the movable amount calculated in the eighth step, and setting the input signal attention window at that position. By repeating the ninth step from the sixth step, An input signal similarity value is calculated for each position of the signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold, whether or not the reference signal is present at the position currently indicated by the input signal attention window A high-speed signal search method characterized by determining whether

4. A first step of generating a feature sequence for a plurality of reference signals registered in advance, and a second process of generating a feature sequence for an input signal.
And a third step of setting an input signal attention window for the feature amount sequence generated in the second step; and for two reference signals of the plurality of reference signals,
A fourth step of calculating a similarity value between reference signals indicating a degree of similarity between the feature quantity sequence for one reference signal and the feature quantity sequence for the other reference signal; and for each reference signal of the plurality of reference signals, First
A fifth step of calculating an input signal similarity value indicating a degree of similarity between the feature amount sequence generated in the step and the feature amount sequence in the input signal attention window; and a reference signal calculated in the fourth step. A sixth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the similarity value and the input signal similarity value calculated in the fifth step; and Determining a position of the input signal attention window based on the movable amount, and setting the input signal attention window at the position, and repeating the fifth to seventh steps. Accordingly, the input signal similarity value is calculated for each position of the input signal attention window, and the reference signal is set at the position indicated by the input signal attention window based on the comparison result between the input signal similarity value and a preset threshold value. High speed characterized by determining whether or not it exists Signal search method.

5. A first step of generating a feature value sequence for a plurality of reference signals registered in advance, and a second step of setting a reference signal attention window for each feature value sequence generated in the first process. A second step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows; and a fourth step of generating a feature amount sequence for the input signal input.
And a fifth step of setting an input signal attention window for the feature amount sequence generated in the fourth step. The input signal attention window corresponds to the plurality of reference signal division attention windows. A sixth step of dividing into a plurality of input signal division windows of interest; and two reference signals of the plurality of reference signals,
A similarity value indicating the degree of similarity between the feature amount sequence related to one reference signal and the feature amount sequence related to the other reference signal, and the feature in each reference signal division attention window corresponding to each other between the two reference signals. A seventh process of calculating a similarity value between reference signals, which is a similarity value indicating a degree of similarity between the quantity sequences, and a feature value sequence in each reference signal division attention window for each of the plurality of reference signals. Eighth step of calculating an input signal similarity value indicating the degree of similarity with the feature amount sequence in the input signal division attention window corresponding to the reference signal division attention window, and reference signal similarity calculated in the seventh step. A ninth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved, based on the value and the input signal similarity value calculated in the eighth step; and a movement calculated in the ninth step. Based on the available Determining a position of the window of interest, and setting the window of interest of the input signal at that position, by repeating the tenth process from the sixth process, Calculating an input signal similarity value for the position and determining whether a reference signal is present at the position indicated by the input signal attention window based on a comparison result between the input signal similarity value and a preset threshold value; A high-speed signal search method characterized by the above-mentioned.

For a plurality of locations wherein said input signal, and the preliminary matching process of obtaining the similarity values between the reference signal and the input signal in advance, a plurality of similarity values obtained by the preliminary verification process, the similarity value A preliminary matching similarity value statistical process for obtaining an average value and a standard deviation value, and a threshold value determining process for determining the threshold value based on the average value and the standard deviation value obtained in the preliminary matching similarity value statistical process. The high-speed signal search method according to any one of claims 1 to 5 , further comprising:

Wherein said input signal similarity value overlaps the high-speed signal searching method according to one of claims 1 to 3, characterized in that a similarity value.

Wherein said input signal of interest window, according to claim 1, 2 or 4 high-speed signal searching method, wherein it is noted window having the same time length as the reference signal in the time axis direction.

Wherein said movable weight, theta is the threshold, S a is the input signal similar value, D is the total frequency of the feature value series, floor {} represents the following devaluation point If

(Equation 1) The high-speed signal search method according to any one of claims 1 to 5 , wherein the high-speed signal search is performed.

Wherein said reference signal and said input signal,
The high-speed signal search method according to any one of claims 3 to 5 , wherein the method is an acoustic signal.

Wherein said input signal of interest window and the reference signal of interest window, according to claim 3 also <br/> 5 characterized in that it is a target window having the same time length as the reference signal in the time axis direction The high-speed signal search method according to the above.

12. The method of claim 11, wherein the reference signal between similar values and the input signal similar values, high-speed signal searching method according to claim 4 or 5, wherein it is overlapped similar value.

13. A first step of generating a feature amount sequence for the reference signal is a previously registered sound signal, and a second step of generating a histogram for the generated feature amount sequence in the first step, the inputted input signal is an acoustic signal, a predetermined time
A third step of setting a long input signal attention window, a fourth step of generating a feature amount sequence for the input signal within the input signal attention window, and a histogram of the feature amount sequence generated in the fourth step. A sixth step of calculating an input signal similarity value indicating a degree of similarity between the histogram created in the second step and the histogram created in the fifth step; Based on the input signal similarity value calculated in the process, a seventh process of calculating a movable amount indicating the amount by which the input signal attention window can be moved; and based on the movable amount calculated in the seventh process, An eighth step of determining the position of the input signal attention window and setting the input signal attention window at the position; and repeating the fourth to eighth steps to obtain the input signal attention window. Input signal for each position Calculating a similarity value, and determining whether or not a reference signal exists at a position currently indicated by the input signal attention window on the input signal based on a comparison result between the input signal similarity value and a preset threshold value. Characteristic high-speed signal search method.

15. advance the registered reference signal and the first process of generating a feature value series, the second process in which the feature amount sequence generated by the first step, to set a reference signal of interest window A third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and a fourth step of generating a feature amount sequence for the input signal input.
And a fifth step of setting an input signal attention window for the feature amount sequence generated in the fourth step. The input signal attention window corresponds to the plurality of reference signal division attention windows. A sixth process of dividing into a plurality of input signal division windows of interest, a seventh process of creating a histogram for a feature sequence in each reference signal division window of interest, and a feature sequence in each input signal division window of interest. Eighth process of creating a histogram, and calculating an input signal similarity value indicating a degree of similarity between the histogram in each reference signal division window and the histogram in the input signal division window corresponding to the reference signal division window. A ninth step, a tenth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the ninth step, and a tenth step Moved Determining an input signal window of interest based on the possible amount, and setting the input signal window of interest at that position; and repeating the eleventh step from the sixth step. The input signal similarity value is calculated for each position of the input signal attention window, and the reference signal is present at the position indicated by the input signal attention window based on the comparison result between the input signal similarity value and a preset threshold value. A high-speed signal search method characterized by determining whether or not to perform a search.

16. A first step of generating a feature sequence for a plurality of reference signals registered in advance, a second process of creating a histogram for the feature sequence generated in the first process, and Generating a feature sequence for the input signal
A fourth step of setting an input signal attention window for the feature amount sequence generated in the third step; and a fifth step of creating a histogram for the feature amount sequence in the input signal attention window. And for two reference signals of the plurality of reference signals,
A sixth step of calculating a similarity value between reference signals indicating a degree of similarity between the histogram related to one reference signal and the histogram related to the other reference signal;
A seventh step of calculating an input signal similarity value indicating a degree of similarity between the histogram generated in the step and the histogram generated in the fifth step; and a similarity value between the reference signals calculated in the sixth step. An eighth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the seventh step; and a movable amount calculated in the eighth step. A ninth step of determining the position of the input signal attention window based on the above, and setting the input signal attention window at that position. By repeating the ninth step from the fifth step, An input signal similarity value is calculated for each position of the input signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold, whether a reference signal is present at the position currently indicated by the input signal attention window. Determining whether or not Fast signal search method.

17. A first step of generating a feature amount sequence for a plurality of reference signals registered in advance, and a second step of setting a reference signal attention window for each feature amount sequence generated in the first step. A second step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows; and a fourth step of generating a feature amount sequence for the input signal input.
And a fifth step of setting an input signal attention window for the feature amount sequence generated in the fourth step. The input signal attention window corresponds to the plurality of reference signal division attention windows. A sixth process of dividing into a plurality of input signal division windows of interest, a seventh process of creating a histogram for a feature sequence in each reference signal division window of interest, and a feature sequence in each input signal division window of interest. An eighth step of creating a histogram; and two reference signals of the plurality of reference signals,
A similarity value indicating the similarity between the histogram for one reference signal and the histogram for the other reference signal, and the similarity between the histograms in the reference signal division attention windows corresponding to each other between the two reference signals. A ninth step of calculating a similarity value between reference signals, which is a similarity value indicating a histogram of each of the plurality of reference signals. A tenth step of calculating an input signal similarity value indicating the degree of similarity with the histogram in the input signal division window of interest, and a similarity value between the reference signals calculated in the ninth step and the tenth step. First calculating a movable amount indicating the amount by which the input signal attention window can be moved based on the input signal similarity value;
And a twelfth step of determining the position of the input signal window of interest based on the movable amount calculated in the eleventh step, and setting the input signal window of interest at that position. By repeating the sixth to twelfth steps, an input signal similarity value is calculated for each position of the input signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold value. And determining whether or not a reference signal is present at a position currently indicated by the input signal attention window.

21. The feature amount sequence is a sequence of the number of zero crossings of a reference signal and an input signal and a derivative thereof, and the histogram includes a plurality of ranges of values that the number of zero crosses and the derivative thereof can take. is divided into sections, one section wherein the claims 13 to 20, characterized in that is created by the feature amount sequence corresponding to each section counted on the basis of the number of zero-crossings and its differential value Fast signal search method.

22. The feature amount sequence is a sequence of feature vectors having components of a plurality of frequency bands and elements, the histogram divides the range of each element can take the value of the feature vectors into a plurality of sections and, wherein claims 13 to 20, characterized in that is created by the feature amount sequence corresponding to each section counted based on the value of the element
13. The high-speed signal search method according to any one of the above items.

23. A reference signal feature amount generating means for generating a feature amount sequence for a reference signal which is a previously registered audio signal, and an input signal feature amount for generating a feature amount sequence for an input signal which is an input audio signal. Generating means, input signal attention window setting means for setting an input signal attention window of a fixed time length for the feature amount sequence generated by the input signal feature amount generation means, and reference signal feature amount generation means. Input signal similarity calculating means for calculating an input signal similarity value indicating the degree of similarity between the feature amount series and the feature amount series in the input signal attention window; and an input signal similarity value calculated by the input signal similarity value calculating means. A movable amount calculating means for calculating a movable amount indicating an amount capable of moving the input signal attention window based on the input signal attention window based on the movable amount calculated by the movable signal amount calculating means. And the input signal attention window position determining means for setting the input signal attention window at that position, the input signal similarity value calculation means, the movable amount calculation means, and the input signal attention window position determination means, respectively. Is repeated to calculate an input signal similarity value for each position of the input signal attention window. Based on a comparison result between the input signal similarity value and a preset threshold value, the input signal A high-speed signal search apparatus comprising: a determination unit configured to determine whether a reference signal is present at a position currently indicated by the input signal attention window.

28. a high-speed signal searching apparatus according to claim 23 or 24, wherein the video device, based on a determination result of the high-speed signal searching device, to a control means for controlling the recording operation of the video device Automatic video control device characterized by the above-mentioned.

The 29. Computer, generates a reference signal feature value generating means for generating a feature amount sequence for the reference signal is a pre-registered audio signal, a feature amount sequence for an input sound signal input signal input A signal feature amount generation unit, an input signal attention window setting unit for setting an input signal attention window of a fixed time length for the feature amount sequence generated by the input signal feature amount generation unit, and a reference signal feature amount generation unit. Input signal similarity value calculating means for calculating an input signal similarity value indicating a degree of similarity between the generated feature amount sequence and the feature amount sequence in the input signal attention window; and an input signal calculated by the input signal similarity value calculating means. A movable amount calculating means for calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the similar value; and an input based on the movable amount calculated by the movable amount calculating means. An input signal attention window position determining means for determining a position of the signal attention window and setting the input signal attention window at the position; an input signal similarity value calculating means; a movable amount calculating means; and an input signal attention window position determining means. For each position of the input signal attention window, a repetition means for calculating the input signal similarity value for each position of the input signal attention window, and a comparison result between the input signal similarity value and a preset threshold value. A recording medium on which is recorded a program for functioning as a judging means for judging whether or not a reference signal is present at a position currently indicated by an input signal attention window on an input signal based on the input signal.

[Procedure amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0007

[Correction method] Change

[Correction contents]

[0007]

According to the first aspect of the present invention,
A first process of generating a feature sequence for a reference signal that is a pre-registered audio signal;
A second step of setting an input signal attention window having a fixed time length for an input signal to be input, a third step of generating a feature amount sequence for an input signal within the input signal attention window, and a first step A fourth step of calculating an input signal similarity value indicating a degree of similarity between the feature quantity sequence generated in step 3 and the feature quantity sequence generated in the third step, and an input signal similarity value calculated in the fourth step And a position of the input signal attention window is determined on the basis of the movable amount calculated in the fifth step. A sixth step of setting the input signal attention window at the position, and repeating the third to sixth steps to obtain an input signal similarity value for each position of the input signal attention window. Is calculated, and the ratio between the input signal similarity value and a preset threshold is calculated. Based on the results, the input signal of interest windows on the input signal is characterized by determining whether the reference signal to the position indicated currently exists. Claim 2
The invention described above refers to a reference which is a video signal of a moving image registered in advance.
A first step of generating a feature sequence for the illumination signal;
The input signal, which is the video signal of the
The second step of setting the input signal attention window of a fixed time length, and
Generate a feature sequence for the input signal within the input signal attention window.
A third process to be performed and a feature amount system generated in the first process
Similarity between the sequence and the feature sequence generated in the third process
A fourth step of calculating an input signal similarity value indicating
Based on the input signal similarity value calculated in the process, the input signal
Fifth calculating the movable amount indicating the amount that the attention window can be moved
And the amount of movement calculated in the fifth step
To determine the position of the input signal attention window, and
And setting a signal attention window.
By repeating the sixth step from the step
Calculate the input signal similarity value for each position of the force signal attention window
And a comparison result between the input signal similarity value and a preset threshold value
Input signal attention window on the input signal based on
To determine whether a reference signal exists at the indicated position.
Features.

[Procedure amendment 3]

[Document name to be amended] Statement

[Correction target item name] 0008

[Correction method] Change

[Correction contents]

According to a third aspect of the present invention, a first step of generating a feature amount sequence for a reference signal registered in advance, and a reference signal attention window for a feature amount sequence generated in the first step are set. A second step of setting, a third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and generating a feature amount sequence for the input signal input A fourth step, a fifth step of setting an input signal attention window for the feature amount series generated in the fourth step, and setting the input signal attention window to the plurality of reference signal division attention windows. A sixth process of dividing into a plurality of corresponding input signal division windows of interest, a feature value sequence in each reference signal division window of interest, and a feature value sequence in an input signal division window of interest corresponding to the reference signal division window of interest. The seventh step of calculating the input signal similarity value indicating the similarity degree of An eighth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the seventh step, and a movable amount calculated in the eighth step Ninth step of determining the position of the input signal attention window based on the quantity and setting the input signal attention window at that position, and repeating the ninth step from the sixth step. The input signal similarity value is calculated for each position of the input signal attention window, and the reference signal is present at the position indicated by the input signal attention window based on the comparison result between the input signal similarity value and a preset threshold value. Is determined.

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0009

[Correction method] Change

[Correction contents]

According to a fourth aspect of the present invention, there is provided a first step of generating a feature sequence for a plurality of reference signals registered in advance, a second process of generating a feature sequence for an input signal input, A third step of setting an input signal attention window with respect to the feature amount sequence generated in the second step; and a feature amount sequence relating to one of the plurality of reference signals. A fourth step of calculating an inter-reference signal similarity value indicating a degree of similarity between the reference signal and the other reference signal, and each of the plurality of reference signals is generated in the first step. A fifth step of calculating an input signal similarity value indicating a degree of similarity between the feature amount series and the feature amount sequence in the input signal attention window; and a fifth step of calculating the similarity value between the reference signals calculated in the fourth step. Input signal similarity value calculated in the process And a position of the input signal attention window is determined on the basis of the movable amount calculated in the sixth step. And a seventh step of setting the input signal attention window at that position, and repeating the fifth to seventh steps to obtain an input signal similarity value for each position of the input signal attention window. Is calculated, based on a comparison result between the input signal similarity value and a preset threshold value,
It is characterized in that it is determined whether or not a reference signal exists at a position currently indicated by the input signal attention window.

[Procedure amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0010

[Correction method] Change

[Correction contents]

According to a fifth aspect of the present invention, a first step of generating a feature amount sequence for a plurality of pre-registered reference signals, and a reference signal generation process for each of the feature amount sequences generated in the first step. A second step of setting an attention window, a third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and a feature amount sequence for the input signal input. A fourth step of generating an input signal attention window for the feature amount sequence generated in the fourth step, and dividing the input signal attention window by the plurality of reference signal divisions. A sixth step of dividing the input signal into windows of interest corresponding to the window of interest;
Each of the reference signal divisions is a similarity value indicating a degree of similarity between a feature value sequence related to one reference signal and a feature value sequence related to the other reference signal, and corresponds to each other between the two reference signals. A seventh process of calculating a similarity value between reference signals, which is a similarity value indicating a degree of similarity between feature amount sequences in the window of interest, and, for each reference signal of the plurality of reference signals, a reference signal division in each window of interest. An input signal similarity value indicating the degree of similarity between the feature amount sequence of the above and the feature amount sequence in the input signal division attention window corresponding to the reference signal division attention window is calculated in an eighth process and a seventh process. A ninth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved, based on the similarity value between the reference signals and the input signal similarity value calculated in the eighth step; Movable calculated by Based on, to determine the position of the input signal of interest window, first setting the input signal of interest window in that position
0, and by repeating the sixth to the tenth steps, an input signal similarity value is calculated for each position of the input signal attention window, and the input signal similarity value is preset as the input signal similarity value. It is characterized in that it is determined whether or not a reference signal exists at a position currently indicated by an input signal attention window based on a result of comparison with a threshold value.

[Procedure amendment 6]

[Document name to be amended] Statement

[Correction target item name] 0011

[Correction method] Change

[Correction contents]

According to a sixth aspect of the present invention, in the high-speed signal search method according to any one of the first to fifth aspects, a similar value between the input signal and the reference signal is determined for a plurality of points of the input signal. A preliminary matching process that is determined in advance, and a plurality of similar values obtained in the preliminary matching process, a preliminary matching similar value statistical process that determines an average value and a standard deviation value of the similar values;
A threshold value determining step of determining the threshold value based on the average value and the standard deviation value obtained in the preliminary matching similarity value statistical step.

[Procedure amendment 7]

[Document name to be amended] Statement

[Correction target item name] 0012

[Correction method] Change

[Correction contents]

[0012] According to a seventh aspect, claim 1-3 Neu
The high-speed signal search method according to any one of the preceding claims, wherein the input signal similarity value is an overlap similarity value.

[Procedure amendment 8]

[Document name to be amended] Statement

[Correction target item name] 0013

[Correction method] Change

[Correction contents]

[0013] The invention according to claim 8 is the invention according to claim 1 , 2 or
5. The high-speed signal search method according to 4 , wherein the input signal attention window is an attention window having the same time length as the reference signal in a time axis direction.

[Procedure amendment 9]

[Document name to be amended] Statement

[Correction target item name] 0014

[Correction method] Change

[Correction contents]

According to a ninth aspect of the present invention, in the high-speed signal search method according to any one of the first to fifth aspects, the movable amount is such that θ is the threshold value and S is the input signal similarity value. And D is the total frequency of the feature amount series, and
When or ｛｝ represents rounding down to the decimal point, it is characterized by being calculated by the equation (1).

[Procedure amendment 10]

[Document name to be amended] Statement

[Correction target item name] 0015

[Correction method] Change

[Correction contents]

According to a tenth aspect of the present invention, in the high-speed signal search method according to any one of the third to fifth aspects, the reference signal and the input signal are acoustic signals.

[Procedure amendment 11]

[Document name to be amended] Statement

[Correction target item name] 0016

[Correction method] Change

[Correction contents]

The invention described in claim 11 is the invention according to claim 3 or 5.
In the high-speed signal search method described above, the input signal attention window and the reference signal attention window are attention windows having the same time length as the reference signal in a time axis direction.

[Procedure amendment 12]

[Document name to be amended] Statement

[Correction target item name] 0017

[Correction method] Change

[Correction contents]

According to a twelfth aspect of the present invention, there is provided the fourth or fifth aspect.
In the high-speed signal search method described above, the similarity value between the reference signals and the similarity value of the input signal are overlapped similarity values.

[Procedure amendment 13]

[Document name to be amended] Statement

[Correction target item name] 0018

[Correction method] Change

[Correction contents]

According to a thirteenth aspect of the present invention, there is provided a method in which
A first step of generating a feature sequence for a reference signal that is an acoustic signal, a second process of creating a histogram for the feature sequence generated in the first process,
A third step of setting an input signal attention window of a fixed time length for an input signal that is an acoustic signal, a fourth step of generating a feature amount sequence for an input signal in the input signal attention window, and A fifth step of creating a histogram for the feature amount series generated in step 4, and an input signal similarity value indicating the degree of similarity between the histogram created in the second step and the histogram created in the fifth step And a seventh step of calculating a movable amount indicating an amount by which the input signal attention window can be moved, based on the input signal similarity value calculated in the sixth step. An eighth step of determining the position of the input signal window of interest based on the movable amount calculated in the step, and setting the input signal window of interest at that position; and Input by repeating the process of 8 For each position of the signal attention window, the input signal similarity value is calculated, and based on the comparison result between the input signal similarity value and a preset threshold, the reference signal is located at the position indicated by the input signal attention window on the input signal. It is characterized in that it is determined whether or not it exists. The invention according to claim 14
Is a reference signal which is a video signal of a moving image registered in advance.
First step of generating a feature quantity sequence for the first and second steps
Create a histogram for the feature value series generated in
A second process, and an input which is a video signal of the input moving image.
Set an input signal attention window of a fixed time length for the force signal
A third process, and an input signal within the input signal attention window.
A fourth step of generating a feature quantity sequence
Create a histogram for the generated feature sequence
The fifth step, the histogram created in the second step,
The degree of similarity with the histogram created in the fifth process
A sixth step of calculating an input signal similarity value
Based on the input signal similarity value calculated in
Calculate the movable amount indicating the amount that the eye window can be moved.
Process and the amount of movement calculated in the seventh process
To determine the position of the input signal attention window, and
An eighth step of setting a signal attention window.
By repeating the eighth step from the step
Calculate the input signal similarity value for each position of the force signal attention window
And a comparison result between the input signal similarity value and a preset threshold value
Input signal attention window on the input signal based on
To determine whether a reference signal exists at the indicated position.
Features.

[Procedure amendment 14]

[Document name to be amended] Statement

[Correction target item name] 0019

[Correction method] Change

[Correction contents]

According to a fifteenth aspect of the present invention, a first step of generating a feature amount sequence for a reference signal registered in advance,
A second step of setting a reference signal attention window for the feature amount sequence generated in the first step, and a reference signal attention window set in the second step as a plurality of reference signal division attention windows. A third step of dividing, a fourth step of generating a feature value sequence for the input signal input, and a fifth step of setting an input signal attention window for the feature value sequence generated in the fourth process. And a sixth step of dividing the input signal attention window into a plurality of input signal division attention windows corresponding to the plurality of reference signal division attention windows, and a feature amount sequence in each reference signal division attention window. A seventh step of creating a histogram, an eighth step of creating a histogram for a feature amount sequence in each input signal division window of interest, and a histogram in each reference signal division window of interest and corresponding to the reference signal division window of interest. In the input signal division window of interest A ninth step of calculating an input signal similarity value indicating the degree of similarity with the togram, and a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the ninth step. Based on a tenth step of calculating and the movable amount calculated in the tenth step,
An eleventh step of determining the position of the input signal attention window and setting the input signal attention window at the position, and repeating the sixth to eleventh steps to obtain the input signal attention window. For each of the positions, the input signal similarity value is calculated, and based on the comparison result between the input signal similarity value and a preset threshold value, it is determined whether or not the reference signal is present at the position currently indicated by the input signal attention window. It is characterized by doing.

[Procedure amendment 15]

[Document name to be amended] Statement

[Correction target item name] 0020

[Correction method] Change

[Correction contents]

According to a sixteenth aspect of the present invention, there is provided a first step of generating a feature sequence for a plurality of reference signals registered in advance, and a second process of generating a histogram for the feature sequence generated in the first process. , A third step of generating a feature amount sequence for the input signal that has been input, a fourth step of setting an input signal attention window for the feature amount sequence generated in the third step, A fifth step of creating a histogram for the feature amount series in the input signal attention window, and, for two reference signals of the plurality of reference signals, a histogram of one reference signal and a histogram of the other reference signal. A sixth step of calculating a similarity value between reference signals indicating a degree of similarity, and a histogram generated in the second step and a fifth step for each of the plurality of reference signals. 7th step of calculating an input signal similarity value indicating the degree of similarity with the histogram generated in the step, and the similarity between the reference signals calculated in the sixth step and the input signal similarity calculated in the seventh step. An eighth step of calculating a movable amount indicating the amount by which the input signal attention window can be moved based on the value and the position of the input signal attention window based on the movable amount calculated in the eighth step. A ninth step of determining and setting the input signal attention window at that position, and repeating the ninth step from the fifth step,
An input signal similarity value is calculated for each position of the input signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold, whether a reference signal is present at the position currently indicated by the input signal attention window. It is characterized in that it is determined whether or not it is.

[Procedure amendment 16]

[Document name to be amended] Statement

[Correction target item name] 0021

[Correction method] Change

[Correction contents]

According to a seventeenth aspect of the present invention, a first step of generating a feature value sequence for a plurality of reference signals registered in advance, and a feature value sequence generated in the first process,
A second step of setting a reference signal attention window, a third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and a feature of the input signal. A fourth step of generating a quantity sequence; a fifth step of setting an input signal attention window for the feature quantity sequence generated in the fourth step; A sixth process of dividing into a plurality of input signal division windows of interest corresponding to the signal division window of interest, a seventh process of creating a histogram for a feature amount sequence in each reference signal division of window of interest, An eighth process of creating a histogram for the feature amount series in the window, and for two reference signals of the plurality of reference signals, determining the similarity between the histogram for one reference signal and the histogram for the other reference signal. A to similar values, and the two
A ninth step of calculating a similarity value between reference signals, which is a similarity value indicating a similarity degree between histograms in the reference signal division attention windows corresponding to each other between one reference signal, and each reference among a plurality of reference signals A tenth step of calculating, for the signal, an input signal similarity value indicating a degree of similarity between the histogram in each reference signal division window of interest and the histogram in the input signal division window of interest corresponding to the reference signal division window of interest;
Calculating a movable amount indicating an amount by which the input signal attention window can be moved, based on the reference signal similarity value calculated in the ninth process and the input signal similarity value calculated in the tenth process;
Step 1 and a twelfth step of determining the position of the input signal attention window based on the movable amount calculated in the eleventh step, and setting the input signal attention window at the position. By repeating the sixth to twelfth steps, an input signal similarity value is calculated for each position of the input signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold value. It is characterized in that it is determined whether or not a reference signal is present at the position currently indicated by the input signal attention window.

[Procedure amendment 17]

[Document name to be amended] Statement

[Correction target item name] 0022

[Correction method] Change

[Correction contents]

The invention according to claim 18 is the invention according to claims 13 to 1
5. The high-speed signal search method according to any one of
The input signal similarity value is an overlap similarity value.
are doing. The invention according to claim 19 is the invention according to claims 13 and 14.
Or the high-speed signal search method according to 16, wherein the input signal
Signal attention window is the same as the reference signal in the time axis direction.
It is a featured window that has an interval. Claim
The invention described in Item 20 is the invention according to any one of Items 15 to 17.
The method of claim 7, wherein the reference signal and
The input signal is an acoustic signal.
You. According to a twenty- first aspect of the present invention, in the high-speed signal search method according to any one of the thirteenth to twentieth aspects, the feature amount sequence is a sequence of a zero crossing number of a reference signal and an input signal and a differential value thereof. The histogram may divide the range of possible values of the number of zero crossings and its differential value into a plurality of sections, and count a feature amount series corresponding to each section based on the number of zero crossings and its differential value. It is characterized by being created by.

[Procedure amendment 18]

[Document name to be amended] Statement

[Correction target item name] 0023

[Correction method] Change

[Correction contents]

The invention according to claim 22 is the invention according to claims 13 to 2.
0 , in the high-speed signal search method according to any one of the above items, the feature amount sequence is a sequence of feature vectors having components of a plurality of frequency bands as elements, and the histogram can take each element of the feature vector. It is characterized by being created by dividing a value range into a plurality of sections and counting a feature amount series corresponding to each section based on the value of the element.

[Procedure amendment 19]

[Document name to be amended] Statement

[Correction target item name] 0024

[Correction method] Change

[Correction contents]

According to the twenty- third aspect of the present invention, there is provided a method in which
The input signal characteristic quantity generating means for generating a reference signal feature value generating means for generating a feature amount sequence, the feature amount sequence for the input acoustic signal <br/> input signal for the reference signal is an acoustic signal, input An input signal attention window setting means for setting an input signal attention window having a fixed time length for the feature value sequence generated by the signal feature value generation means; and a feature value sequence generated by the reference signal feature value generation means. Input signal similarity calculating means for calculating an input signal similarity value indicating a degree of similarity with the feature amount series in the input signal attention window; and an input signal similarity value calculated by the input signal similarity value calculating means. A movable amount calculating means for calculating a movable amount indicating an amount capable of moving the attention window, and a position of the input signal attention window is determined based on the movable amount calculated by the movable amount calculating means; The input signal The input signal attention window position determination means for setting the attention window, the input signal similarity value calculation means, the movable amount calculation means, and the input signal attention window position determination means are made to repeatedly perform the respective processes, thereby obtaining the input. A repetition means for calculating an input signal similarity value for each position of the signal attention window, and a position indicated by the input signal attention window on the input signal based on a comparison result between the input signal similarity value and a preset threshold value And determining means for determining whether or not a reference signal exists. Claim 24
The described invention is a video signal of a moving image registered in advance.
Reference signal features that generate feature sequences for reference signals
Generating means, and an input signal which is a video signal of an input moving image.
Input signal feature generator that generates a feature sequence for a signal
Stage and a feature value sequence generated by the input signal feature value generation means
To set an input signal attention window of a certain time length
Generated by signal attention window setting means and reference signal feature quantity generation means
Feature sequence and feature sequence in the input signal attention window
An input signal that calculates an input signal similarity value indicating the degree of similarity with
Signal similarity value calculation means and input signal similarity value calculation means.
Moves input signal attention window based on input signal similarity value
Moveable amount calculator that calculates the amount of movement that indicates the amount that can be done
Based on the steps and the movable amount calculated by the movable amount calculating means.
Then, the position of the input signal attention window is determined, and
Input signal attention window position determining means for setting an input signal attention window
And input signal similarity value calculating means and movable amount calculating means.
Each processing is performed for the force signal attention window position determination means.
Is repeated, so that each of the input signal attention windows
Repeating means for calculating input signal similarity values for positions
And the comparison result between the input signal similarity value and a preset threshold
Input signal attention window on the input signal based on
A determination step for determining whether or not a reference signal exists at the indicated position
And a step. Claim 25
A high-speed signal search apparatus according to claim 23 or 24.
The characteristic generated by the reference signal characteristic amount generating means.
First histogram for creating a histogram for the collection series
Generated by the input signal feature quantity generating means.
The input signal attention window setting means in the extracted feature amount sequence.
Of the input signal attention window set in
A second histogram creating means for creating a gram.
Wherein the input signal similarity value calculating means includes the first hiss.
The histogram created by the togram creating means and the second
With the histogram created by the histogram creation means
A similarity value indicating the degree of similarity is defined as the input signal similarity value.
It is characterized by calculating. The invention according to claim 26
The high-speed signal search apparatus according to claim 25, wherein
The feature sequence is based on the number of zero crossings of the reference signal and the input signal.
And a series of differential values of the first and second histograms.
The ram generating means respectively includes the number of zero crossings and the fineness thereof.
Divide the range of values that can be taken into a plurality of sections,
The feature amount series corresponding to the section is calculated by
Create a histogram by counting based on the fractional values
It is characterized by The invention according to claim 27
The high-speed signal search apparatus according to claim 25, wherein
A feature sequence is a feature sequence that includes components from multiple frequency bands.
A series of feature vectors, wherein the first and second histograms are
Each of the elements of the feature vector is
Divide the range of values that can be taken into multiple sections, and
Count the corresponding feature value series based on the value of the element
Is characterized by creating a histogram by
I have.

[Procedure amendment 20]

[Document name to be amended] Statement

[Correction target item name] 0025

[Correction method] Change

[Correction contents]

The invention according to claim 28 is the invention according to claim 23 or
24. A high-speed signal search device according to claim 24 , a video device, and control means for controlling a recording operation of the video device based on a determination result of the high-speed signal search device.

[Procedure amendment 21]

[Document name to be amended] Statement

[Correction target item name] 0026

[Correction method] Change

[Correction contents]

According to a twenty- ninth aspect of the present invention, there is provided a computer comprising: a reference signal generation unit configured to generate a characteristic amount sequence for a reference signal that is a pre-registered audio signal; Input signal feature value generation means for generating a quantity sequence; input signal attention window setting means for setting an input signal attention window of a fixed time length for the feature value sequence generated by the input signal feature value generation means; Input signal similarity calculation means for calculating an input signal similarity value indicating a degree of similarity between the feature quantity sequence generated by the signal feature quantity generation means and the feature quantity sequence in the input signal attention window, and input signal similarity value calculation means A movable amount calculating means for calculating a movable amount indicating an amount capable of moving the input signal attention window based on the input signal similarity value calculated in the step; and a movable amount calculated by the movable amount calculating means. Based on the position of the input signal attention window, the input signal attention window position determination means for setting the input signal attention window at the position, the input signal similarity value calculation means, the movable amount calculation means, and the input signal attention window. A repetition means for calculating the input signal similarity value for each position of the input signal attention window by causing the window position determination means to repeatedly perform the respective processes; and an input signal similarity value and a preset threshold value. And a program for functioning as a judging means for judging whether or not a reference signal is present at a position currently indicated by the input signal attention window on the input signal based on a comparison result with the recording medium. Claim 30
Akira uses a computer to send pre-registered video images.
Signal that generates a feature sequence for the reference signal
Signal feature amount generation means, and a video signal of the input moving image.
Signal features that generate a feature sequence for the input signal
Quantity generating means and the characteristic generated by the input signal feature quantity generating means.
Set input signal attention window of fixed time length for collection series
Input signal attention window setting means, and a reference signal feature amount generating means.
The characteristic amount sequence generated in the step and the characteristic in the input signal attention window
Calculate the input signal similarity value indicating the degree of similarity with the collection series
Input signal similarity value calculating means, and input signal similarity value calculating means
Note the input signal based on the input signal similarity value calculated in
Calculate the amount of movement that indicates how much you can move the window
The amount calculated by the amount calculating means and the movable amount calculated by the movable amount calculating means
The position of the input signal attention window is determined based on the
Input signal attention window position for setting the input signal attention window at the position
Determination means, input signal similarity value calculation means and movable amount calculation
Means and the input signal attention window position determining means, respectively.
By repeating this process, the input signal
For each position of the eye window, iterative calculation of the input signal similarity value
Return means, and a ratio between the input signal similarity value and a preset threshold value
Input signal attention window on the input signal based on the comparison result
Determines whether there is a reference signal at the position indicated by
Recorded a program to function as judgment means
It is a recording medium. The invention according to claim 31 is the invention according to claim 29
31. The recording medium according to claim 30, wherein the reference signal feature quantity
Histogram for the feature sequence generated by the generation means
First histogram creating means for creating the input signal;
Out of the feature amount series generated by the signal feature amount generation means,
Within the input signal attention window set by the input signal attention window setting means
Second histogram to create a histogram for
If you let the computer function as a program
In both cases, the input signal similarity value calculating means outputs the first hiss.
The histogram created by the togram creating means and the second
With the histogram created by the histogram creation means
A similarity value indicating the degree of similarity is defined as the input signal similarity value.
To make a computer function as a calculator
This is a recording medium on which a program is recorded. Claim 32
The invention according to claim 31 is the recording medium according to claim 31, wherein
The collection series is based on the number of zero crossings of the reference signal and the input signal and
A series of differential values of the first and second histograms;
Means for generating the zero crossing and the derivative thereof
The range of possible values is divided into multiple sections,
The number of zero crossings and its derivative
Create histograms by counting based on values
To make the computer work
This is a recording medium on which a gram is recorded. Claim 33
32. The recording medium according to claim 31, wherein the characteristic amount
A sequence is a feature vector that includes components from multiple frequency bands.
The first and second histograms
Means for taking each element of said feature vector, respectively
Value range is divided into a plurality of sections, and
Counting the feature value series based on the value of the element.
Therefore, computer
On a recording medium on which a program for operating the
is there.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Gavin Smith 51 Acrefield Drive Cambridge Cambridgeshire United Kingdom F-term (reference) 5C053 FA23 FA27 FA30 GB09 HA29 JA01 JA30 KA14 KA24 LA11 LA14 5D015 AA04 AA06 HH04 KK01 9A001 FF03 GG05 GG14 HH15

Claims (20)

[Claims] A first step of generating a feature amount sequence for a pre-registered reference signal; a second step of setting an input signal attention window for an input signal that has been input; A third step of generating a feature sequence for the input signal in the window; and an input signal similarity indicating a degree of similarity between the feature sequence generated in the first process and the feature sequence generated in the third process. A fourth step of calculating a value, a fifth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the fourth step, Determining the position of the input signal attention window based on the movable amount calculated in the step, and setting the input signal attention window at that position. By repeating the sixth process, each position of the input signal attention window is The input signal similarity value is calculated for the input signal, and based on a comparison result between the input signal similarity value and a preset threshold, whether or not the reference signal is present at the position indicated by the input signal attention window on the input signal. A high-speed signal search method, characterized in that: 2. A first step of generating a feature sequence for a reference signal registered in advance, and a second process of setting a reference signal attention window for the feature sequence generated in the first process. A third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and a fourth step of generating a feature amount sequence for the input signal input.
And a fifth step of setting an input signal attention window for the feature amount sequence generated in the fourth step. The input signal attention window corresponds to the plurality of reference signal division attention windows. A sixth process of dividing into a plurality of input signal division windows of interest, and a similarity between a feature sequence in each reference signal division window of interest and a feature sequence in the input signal division window of interest corresponding to the reference signal division window of interest. A seventh step of calculating an input signal similarity value indicating the degree; and an eighth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the seventh step. And a ninth step of determining the position of the input signal attention window based on the movable amount calculated in the eighth step, and setting the input signal attention window at that position. By repeating the ninth step from the sixth step, An input signal similarity value is calculated for each position of the signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold, whether or not the reference signal is present at the position currently indicated by the input signal attention window A high-speed signal search method characterized by determining whether 3. A first step of generating a feature sequence for a plurality of reference signals registered in advance, and a second process of generating a feature sequence for an input signal.
And a third step of setting an input signal attention window for the feature amount sequence generated in the second step; and for two reference signals of the plurality of reference signals,
A fourth step of calculating a similarity value between reference signals indicating a degree of similarity between the feature quantity sequence for one reference signal and the feature quantity sequence for the other reference signal; and for each reference signal of the plurality of reference signals, First
A fifth step of calculating an input signal similarity value indicating a degree of similarity between the feature amount sequence generated in the step and the feature amount sequence in the input signal attention window; and a reference signal calculated in the fourth step. A sixth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the similarity value and the input signal similarity value calculated in the fifth step; and Determining a position of the input signal attention window based on the movable amount, and setting the input signal attention window at the position, and repeating the fifth to seventh steps. Accordingly, the input signal similarity value is calculated for each position of the input signal attention window, and the reference signal is set at the position indicated by the input signal attention window based on the comparison result between the input signal similarity value and a preset threshold value. High speed characterized by determining whether or not it exists Signal search method. 4. A first step of generating a feature amount sequence for a plurality of reference signals registered in advance, and a second step of setting a reference signal attention window for each feature amount sequence generated in the first step. A second step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows; and a fourth step of generating a feature amount sequence for the input signal input.
And a fifth step of setting an input signal attention window for the feature amount sequence generated in the fourth step. The input signal attention window corresponds to the plurality of reference signal division attention windows. A sixth step of dividing into a plurality of input signal division windows of interest; and two reference signals of the plurality of reference signals,
A similarity value indicating the degree of similarity between the feature amount sequence related to one reference signal and the feature amount sequence related to the other reference signal, and the feature in each reference signal division attention window corresponding to each other between the two reference signals. A seventh process of calculating a similarity value between reference signals, which is a similarity value indicating a degree of similarity between the quantity sequences, and a feature value sequence in each reference signal division attention window for each of the plurality of reference signals. Eighth step of calculating an input signal similarity value indicating the degree of similarity with the feature amount sequence in the input signal division attention window corresponding to the reference signal division attention window, and reference signal similarity calculated in the seventh step. A ninth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved, based on the value and the input signal similarity value calculated in the eighth step; and a movement calculated in the ninth step. Based on the available Determining a position of the window of interest, and setting the window of interest of the input signal at that position, by repeating the tenth process from the sixth process, Calculating an input signal similarity value for the position and determining whether a reference signal is present at the position indicated by the input signal attention window based on a comparison result between the input signal similarity value and a preset threshold value; A high-speed signal search method characterized by the above-mentioned. 5. A preliminary matching step in which similarity between the input signal and the reference signal is determined in advance for a plurality of locations of the input signal, and the similarity value is determined for a plurality of similar values obtained in the preliminary matching step. A preliminary matching similarity value statistical process for obtaining an average value and a standard deviation value, and a threshold value determining process for determining the threshold value based on the average value and the standard deviation value obtained in the preliminary matching similarity value statistical process. The high-speed signal search method according to any one of claims 1 to 4, further comprising: 6. The high-speed signal search method according to claim 1, wherein the input signal similar value is an overlap similar value. 7. The high-speed signal search method according to claim 1, wherein the input signal attention window is an attention window having the same time length as the reference signal in a time axis direction. 8. The movable amount, θ is the threshold value, S is the input signal similarity value, D is the total frequency of the feature amount sequence, and floor ｛｝ represents a round-down below a decimal point. In the case, 5. The high-speed signal search method according to claim 1, wherein the high-speed signal search is performed by: 9. The high-speed signal search method according to claim 1, wherein the reference signal and the input signal are sound signals. 10. The high-speed signal search according to claim 2, wherein the input signal attention window and the reference signal attention window are attention windows having the same time length as the reference signal in a time axis direction. Method. 11. The high-speed signal search method according to claim 3, wherein the inter-reference signal similarity value and the input signal similarity value are overlapping similarity values. 12. A first step of generating a feature sequence for a reference signal registered in advance, a second process of creating a histogram for the feature sequence generated in the first process, A third step of setting an input signal attention window for the signal; a fourth step of generating a feature value sequence for the input signal in the input signal attention window; and a feature value generated in the fourth step A fifth step of creating a histogram for the series, a sixth step of calculating an input signal similarity value indicating a degree of similarity between the histogram created in the second step and the histogram created in the fifth step, A seventh step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the sixth step, and a movable amount calculated in the seventh step. Based on input An eighth step of determining the position of the signal attention window and setting the input signal attention window at that position. By repeating the fourth to eighth steps, An input signal similarity value is calculated for each position, and based on a comparison result between the input signal similarity value and a preset threshold, whether or not a reference signal is present at a position indicated by the input signal attention window on the input signal. A high-speed signal search method characterized by determining whether 13. A first step of generating a feature sequence for a reference signal registered in advance, and a second process of setting a reference signal attention window for the feature sequence generated in the first process. A third step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows, and a fourth step of generating a feature amount sequence for the input signal input.
And a fifth step of setting an input signal attention window for the feature amount sequence generated in the fourth step. The input signal attention window corresponds to the plurality of reference signal division attention windows. A sixth process of dividing into a plurality of input signal division windows of interest, a seventh process of creating a histogram for a feature sequence in each reference signal division window of interest, and a feature sequence in each input signal division window of interest. Eighth process of creating a histogram, and calculating an input signal similarity value indicating a degree of similarity between the histogram in each reference signal division window and the histogram in the input signal division window corresponding to the reference signal division window. A ninth step, a tenth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the ninth step, and a tenth step Moved Determining an input signal window of interest based on the possible amount, and setting the input signal window of interest at that position; and repeating the eleventh step from the sixth step. The input signal similarity value is calculated for each position of the input signal attention window, and the reference signal is present at the position indicated by the input signal attention window based on the comparison result between the input signal similarity value and a preset threshold value. A high-speed signal search method characterized by determining whether or not to perform a search. 14. A first step of generating a feature sequence for a plurality of reference signals registered in advance, a second process of creating a histogram for the feature sequence generated in the first process, and Generating a feature sequence for the input signal
A fourth step of setting an input signal attention window for the feature amount sequence generated in the third step; and a fifth step of creating a histogram for the feature amount sequence in the input signal attention window. And for two reference signals of the plurality of reference signals,
A sixth step of calculating a similarity value between reference signals indicating a degree of similarity between the histogram related to one reference signal and the histogram related to the other reference signal;
A seventh step of calculating an input signal similarity value indicating a degree of similarity between the histogram generated in the step and the histogram generated in the fifth step; and a similarity value between the reference signals calculated in the sixth step. An eighth step of calculating a movable amount indicating an amount by which the input signal attention window can be moved based on the input signal similarity value calculated in the seventh step; and a movable amount calculated in the eighth step. A ninth step of determining the position of the input signal attention window based on the above, and setting the input signal attention window at that position. By repeating the ninth step from the fifth step, An input signal similarity value is calculated for each position of the input signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold, whether a reference signal exists at the position currently indicated by the input signal attention window. Determining whether or not Fast signal search method. 15. A first step of generating a feature value sequence for a plurality of reference signals registered in advance, and a second step of setting a reference signal attention window for each feature value sequence generated in the first process. A second step of dividing the reference signal attention window set in the second step into a plurality of reference signal division attention windows; and a fourth step of generating a feature amount sequence for the input signal input.
And a fifth step of setting an input signal attention window for the feature amount sequence generated in the fourth step. The input signal attention window corresponds to the plurality of reference signal division attention windows. A sixth process of dividing into a plurality of input signal division windows of interest, a seventh process of creating a histogram for a feature sequence in each reference signal division window of interest, and a feature sequence in each input signal division window of interest. An eighth step of creating a histogram, and for two reference signals of the plurality of reference signals,
A similarity value indicating the similarity between the histogram for one reference signal and the histogram for the other reference signal, and the similarity between the histograms in the reference signal division attention windows corresponding to each other between the two reference signals. A ninth step of calculating a similarity value between reference signals, which is a similarity value indicating a histogram of each of the plurality of reference signals. A tenth step of calculating an input signal similarity value indicating the degree of similarity with the histogram in the input signal division window of interest, and a similarity value between the reference signals calculated in the ninth step and the tenth step. First calculating a movable amount indicating the amount by which the input signal attention window can be moved based on the input signal similarity value;
And a twelfth step of determining the position of the input signal window of interest based on the movable amount calculated in the eleventh step, and setting the input signal window of interest at that position. By repeating the sixth step to the twelfth step, an input signal similarity value is calculated for each position of the input signal attention window, and based on a comparison result between the input signal similarity value and a preset threshold value. A method for determining whether a reference signal is present at a position currently indicated by an input signal attention window. 16. The feature amount sequence is a sequence of a number of zero crossings of a reference signal and an input signal and a derivative thereof, and the histogram includes a plurality of ranges of values that the number of zero crossings and a derivative thereof can take. 16. The method according to any one of claims 12 to 15, wherein the area is created by dividing the section into sections and counting a feature amount sequence corresponding to each section based on the number of zero crossings and a derivative thereof. Fast signal search method. 17. The feature amount sequence is a sequence of feature vectors having components of a plurality of frequency bands as elements, and the histogram divides a range of values that each element of the feature vector can take into a plurality of sections. The feature amount sequence is created by counting a feature amount sequence corresponding to each section based on the value of the element.
13. The high-speed signal search method according to any one of the above items. 18. A reference signal feature value generating means for generating a feature value sequence for a reference signal registered in advance, an input signal feature value generating means for generating a feature value sequence for an input signal inputted, and an input signal feature value. Input signal attention window setting means for setting an input signal attention window with respect to the feature value sequence generated by the generation means; feature value sequences generated by the reference signal feature value generation means; and features in the input signal attention window. Input signal similarity value calculating means for calculating an input signal similarity value indicating the degree of similarity with the quantity series; and an amount by which the input signal attention window can be moved based on the input signal similarity value calculated by the input signal similarity value calculating means. A movable amount calculating means for calculating the movable amount shown, and a position of the input signal attention window is determined based on the movable amount calculated by the movable amount calculating means, and the input signal attention window is set at the position. The input signal attention window position determination means, the input signal similarity value calculation means, the movable amount calculation means, and the input signal attention window position determination means are repeatedly subjected to the respective processes, whereby the input signal attention window A repetition means for calculating an input signal similarity value for each position; and a reference signal at a position currently indicated by the input signal attention window on the input signal based on a comparison result between the input signal similarity value and a preset threshold. A high-speed signal search device comprising: a determination unit configured to determine whether the signal exists. 19. A high-speed signal search device according to claim 18, comprising: a video device; and control means for controlling a recording operation of the video device based on a determination result of the high-speed signal search device. Video automatic control device. 20. A computer, comprising: a reference signal feature quantity generating means for generating a feature quantity sequence for a reference signal registered in advance; an input signal feature quantity generating means for generating a feature quantity sequence for an input signal input; Input signal attention window setting means for setting an input signal attention window with respect to the feature value sequence generated by the signal feature value generation means; a feature value sequence generated by the reference signal feature value generation means; Input signal similarity calculating means for calculating an input signal similarity value indicating the degree of similarity with the feature amount series in the input signal similarity value calculating means, and moving the input signal attention window based on the input signal similarity value calculated by the input signal similarity value calculating means A movable amount calculating means for calculating a movable amount indicating a possible amount; and a position of the input signal attention window is determined based on the movable amount calculated by the movable amount calculating means. The input signal attention window position determination means for setting the force signal attention window, the input signal similarity value calculation means, the movable amount calculation means, and the input signal attention window position determination means are made to repeat the respective processes. An iterative means for calculating the input signal similarity value for each position of the input signal attention window; and based on the comparison result between the input signal similarity value and a preset threshold, the input signal attention window is A recording medium on which a program for functioning as a judging means for judging whether or not a reference signal is present at the indicated position is recorded.