JP2003108185A - Time-series signal analyzing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a time-series signal analyzing device which can take a more accurate analysis by taking cyclic variation of a heat-sound signal into consideration. SOLUTION: A frequency analyzing means 2 generates a plurality of pieces of phoneme data composed of a pair of a frequency and intensity by taking a frequency analysis of the heart sound signal and a block phoneme generating means 3 generates a block phoneme by using the generated phoneme data. A fundamental cycle calculating means 4, on the other hand, calculates the fundamental cycles of the heart sound signal by taking an autocorrelation analysis of the heart sound signal. A block phoneme classifying means 5 classifies block phonemes which have similar frequencies and differences in start time close to integral multiples of the fundamental cycles into the same group. The classified block phonemes are outputted, group by group, to an output means 9.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technique for analyzing time-series signals including heart sounds, biological signals such as electrocardiogram, and other acoustic signals.

[0002]

2. Description of the Related Art A time series signal represented by an acoustic signal is
A plurality of periodic signals are included as its constituent elements. Therefore, a method of analyzing what kind of periodic signal is included in a given time series signal has been known for a long time. For example, Fourier analysis is widely used as a method for analyzing frequency components included in a given time series signal.

By using such a time-series signal analysis method, it is possible to encode an acoustic signal. With the spread of computers, it has become easy to sample the analog sound signal that is the original sound at a predetermined sampling frequency, quantize the signal strength at each sampling, and capture it as digital data. If a method such as Fourier analysis is applied to the data and the frequency components included in the original sound signal are extracted, the original sound signal can be encoded by the code indicating each frequency component.

On the other hand, MIDI (Musical Instrume) was born from the idea of encoding musical instrument sounds by electronic musical instruments.
The nt Digital Interface) standard has also been actively used with the spread of personal computers. Code data according to this MIDI standard (hereinafter referred to as MID
Basically, the I data) is data that describes the operation of the musical instrument playing, such as which keyboard key of the musical instrument was played and with what strength. The MIDI data itself contains the actual sound. Waveform not included. Therefore, when reproducing the actual sound, the MI that stores the waveform of the instrument sound is stored.
Although a DI sound source is required separately, its high coding efficiency has been attracting attention, and the MIDI coding and decoding technology is currently used for musical instrument performance, musical instrument practice, composition, etc. using a personal computer. It is widely adopted in software that does.

Therefore, a time-series signal typified by an acoustic signal is analyzed by a predetermined method to extract a periodic signal which is a constituent element thereof, and the extracted periodic signal is M
Proposals have been made to encode using IDI data. For example, JP-A-10-247099, JP-A-11-73199, and JP-A-11-73200.
JP, JP-A-11-95753, JP, 2000
-99009, JP 2000-99092 A, JP 2000-99093 A, JP 2000-
261322, JP 2001-5450 A,
Japanese Unexamined Patent Application Publication No. 2001-148633 proposes various methods capable of analyzing a frequency as a constituent element of an arbitrary time series signal and creating MIDI data from the analysis result.

[0006]

In particular, Japanese Patent Application No. 2000-
In the 351775 specification, a method of converting a heart sound into a MIDI code, dividing the obtained MIDI code into blocks, and assigning a heart sound attribute to each block was proposed.
In this method, a unified frequency element at the time is calculated based on a plurality of simultaneously generated frequency elements, the unified frequency elements continuous in time series are divided into blocks, and a heart sound attribute is assigned to this block data. Therefore, there is a problem that an incorrect heart sound attribute may be assigned due to periodic fluctuations of heart sounds.

In view of the above points, it is an object of the present invention to provide a time-series signal analysis device capable of performing a more accurate analysis in consideration of periodic fluctuations of a heart sound signal. .

[0008]

In order to solve the above problems, the present invention provides, as a time-series signal analysis device, a set of frequency and intensity for each predetermined unit section set for a given time-series signal. A frequency analysis unit that generates a plurality of phoneme data, and a plurality of the generated phoneme data that satisfies a predetermined strength is extracted, and the start time and end time of the extracted phoneme data are extracted.・ A blocking means for generating block phoneme data composed of parameters of start time, end time, frequency, and intensity based on frequency similarity, and by performing autocorrelation analysis on time series signals, The basic period analysis means for calculating the basic period of the time-series signal is similar in frequency among the plurality of block phonemes, and the difference in start time is close to an integral multiple of the calculated basic period. Things and wherein the grouping means for classifying to be the same group, the structure and the fact that a display means for displaying the block phonemes classified groups.

According to the present invention, block data is created based on the similarity of a plurality of phoneme data obtained by analyzing a time-series signal, while auto-correlation analysis is performed on the time-series signal to obtain a time-series signal. The signal cycle is calculated, and the created blocks that are separated by the cycle interval are grouped into the same group and all block data is classified into multiple blocks for display. Can easily determine the heart sound attribute. Furthermore, by using the knowledge database in which the heart sound waveform is registered in association with the heart sound attribute including the characteristics and cycle of each block, it is possible to automatically add the heart sound attribute.

[0010]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below in detail with reference to the drawings. (1. Basic principle of time-series signal analysis) First, the basic principle of frequency analysis, which is partially used for analyzing time-series signals, will be described. Since this basic principle is disclosed in each of the above-mentioned publications or specifications, only the outline thereof will be briefly described here. As shown in the upper part of FIG. 1, it is assumed that an analog acoustic signal is given as a time series intensity signal. In the illustrated example, the horizontal axis is the time axis t, and the vertical axis is the signal strength A.
Is taken to show this acoustic signal. In the present invention, first, a process of capturing this analog acoustic signal as a digital acoustic signal is performed. This can be performed by using a conventional general PCM method, sampling the analog acoustic signal at a predetermined sampling frequency, and converting the signal strength A into digital data using a predetermined number of quantization bits. . Here, for convenience of explanation, the waveform of the audio signal digitized by the PCM method is also shown as the same waveform as the analog audio signal in the upper part of FIG.

Next, a plurality of unit sections are set on the time axis t of this digital audio signal. In the illustrated example, six unit sections U1 to U6 are set. The position and length of the i-th unit section Ui on the time axis t are indicated by the coordinate values of the start end si and the end point ei on the time axis t. For example, the unit section U1 is a section having a length of (e1-s1) from the start end s1 to the end e1.

When a plurality of unit sections are set in this way, a predetermined representative frequency and a representative intensity representative of each unit section are defined based on the acoustic signal in each unit section. Here, for the i-th unit section Ui,
A state in which the representative frequency Fi and the representative strength Ai are defined is shown. For example, the representative frequency F1 and the representative intensity A1 are defined for the first unit section U1. The representative frequency F1 is a representative value of the frequency component of the acoustic signal included in the section from the start end s1 to the end e1, and the representative intensity Ai is the acoustic signal included in the section from the start end s1 to the end e1 as well. Is a representative value of the signal strength of. The frequency component contained in the acoustic signal in the unit section U1 is not usually single, but the signal strength generally fluctuates. The point of the basic principle, which is also common to the conventional methods, is that one or more representative frequencies and representative intensities are defined for one unit section, and encoding is performed using these representative values.

That is, when the representative frequency and the representative intensity are defined for each unit section, information indicating the start and end positions of each unit section on the time axis t and the defined representative frequency and representative The code data is generated from the information indicating the intensity, and the acoustic signal of each unit section is expressed by each code data. As a method of encoding an event in which an acoustic signal having a single frequency and a single signal strength lasts for a predetermined period, encoding based on the MIDI standard can be used. Code data according to the MIDI standard (MI
The DI data) can be said to be data representing sounds by musical notes, and in FIG. 1, the concept of code data finally obtained is shown by the musical notes shown in the lower stage.

In the end, the acoustic signal in each unit section includes the pitch information (note number in the MIDI standard) corresponding to the representative frequency F1, the strength information (velocity in the MIDI standard) corresponding to the representative strength A1, and the unit section. Will be converted into code data having length information (delta time in the MIDI standard) corresponding to the length (e1-s1). The information amount of the coded data obtained in this way is significantly smaller than the information amount of the original audio signal, and a dramatic coding efficiency can be obtained. Note that the representative frequency is the function f of the note number n.
Assuming (n), the relationship between the representative frequency f (n) and the note number n (0 ≦ n ≦ 127) is defined by the following [Formula 1].

[Formula 1] f (n) = 440 × 2 ^{γ (n)} γ (n) = (n−69) / 12

Further, the frequency and the cycle are inversely related to each other. If one is determined, the other is also uniquely determined, and if the frequency is determined by the above [Formula 1], the note number is also determined. Therefore, in the present invention, even when any one of the terms frequency, period, and note number is used, the other two terms will be simultaneously determined.

As a concrete method for calculating a set of representative frequencies and intensities for each unit section as described above,
Well-known methods such as short-time Fourier analysis, generalized harmonic analysis method, and zero crossing detection method can be used.

(2. Time-series signal analysis method according to the present invention)
Next, a specific method of time-series signal analysis according to the present invention will be described. FIG. 2 is a flow chart showing an outline of analysis of a time series signal according to the present invention. As shown in FIG. 2, two types of processing are performed on the digital audio signal captured in the PCM format. First, frequency analysis of the time-series signal is performed (step S1). Here, FIG. 3 shows an example of the electrocardiographic waveform captured as a time-series signal. In step S1, the sonic waveform (signal) as shown in FIG.
On the other hand, a unit section is set as described with reference to FIG. 1 in the above basic principle, and a plurality of frequency and intensity groups are calculated for each unit section.

As a specific method for calculating the set of frequency and intensity, the short-time Fourier transform, generalized harmonic analysis, and zero-crossing point detection method can be used as described above. In the present invention, however, the zero-crossing point detection method is used. It is effective to apply. Especially in the frequency analysis in step S1,
The method of developing the zero crossing detection method is used to calculate the set of frequency and intensity. Details of the frequency analysis in step S1 will be described with reference to the flowchart of FIG.

First, the main zero crossing point is detected from the input original acoustic signal from the signal (step S11). The main zero-crossing point indicates a zero-crossing point, which is a point where the acoustic signal becomes 0 level, that is detected for the original acoustic signal that has not been corrected. Here, a part of the original acoustic signal shown in FIG. 3 is enlarged in the time axis direction, as shown in FIG.
It shows in (a). In step S11, a zero crossing point at which the signal becomes 0 level is detected. This zero crossing is
It will be expressed in time. The main zero crossings detected from the original acoustic signal of FIG. 5 (a) are shown in FIG. 5 (b).

Next, a unit section is determined according to the main zero crossing point detected in step S11 (step S12). The unit interval is a basic interval for extracting frequency components and creating code data, and corresponds to the unit interval U in FIG. The unit section is set as a section from the main zero crossing point t1 which is the start point to the main zero crossing point t2 which is the end point. As the main zero-crossing point t1 that is the starting point, the main zero-crossing point that is temporally leading in the section that has not yet been set as a unit section is selected. The main zero crossing t2, which is the end point, is selected as follows.

As shown in FIG. 5 (a), the waveform of the original acoustic signal has one cycle at the zero crossing point at a position two points away from the starting zero crossing point. Therefore, it is preferable to set the end point at the position where the zero-crossing point that is an integral multiple is moved.
Further, in order to obtain a stable frequency value, it is preferable to set the end point as far away from the start point as possible and set the average period as the period of this unit section. For that purpose, first, the distance from the main zero-crossing point t1 which is the starting point to the main zero-crossing point moved by two is set as the unit cycle T _B. As the main zero-crossing point t2 that is the end point, among the main zero-crossing points that are separated by an integer multiple of 2 from the main zero-crossing point t1 that is the starting point, the one that maximizes t2-t1 within the range that satisfies the following [Formula 2]. Try to be elected.

[0023] [Equation 2] | N _A -N _B | ≦ 1, _{however, N = 40 × log 10 (} 1 / 440T) +69 T = (t2-t1) × 2 / k

In the above [Formula 2], N _A is t2-t
Note number determined by the average period T _A between 1, N _B is the note number determined by the unit cycle T _B. Further, k is the number of main zero crossings moved from the main zero crossing as the start point to the main zero crossing as the end point, and is an integer multiple of 2. In the above [Formula 2],
The second formula is for obtaining the note number N from the period T, and the third formula is for obtaining the average period T _{A. When} k = 2, the unit period T _B is obtained. . In the above [Formula 2], four main zero-crossing points t2, which are the end points, from the main zero-crossing point t1, which is the starting point,
It is temporarily set at the main zero crossing points that are separated by an integral multiple of 2, such as 6 pieces, 8 pieces, and so on. After all, when the frequency determined by the unit period T _B and the frequency determined by the average period T _A are converted into note numbers, the process in step S12 falls within a semitone difference (± 1 difference as a note number). A process of selecting such a main zero crossing t2 as an end point is performed. The unit section is set over the entire section of the acoustic signal. In the subsequent unit section, the main zero crossing next to the end point of the preceding unit section, which is the main zero crossing next to the unit zero, is set as the main zero crossing as the start point, and the same process as described above is performed.

After setting all unit sections, the main frequency component is calculated for each unit section (step S1).
3). The frequency component means a period (frequency) and an amplitude. The average period T _A in the unit section is set as the cycle, and the value A at which the absolute value of the signal is maximum is set as the amplitude in the unit section. The main frequency component refers to a frequency component extracted from the original acoustic signal that has not been corrected, and is distinguished from the sub frequency component calculated in step S16.

Steps S12 and S13 described above
The process according to (1), that is, the process of determining the unit section using the zero crossing point and calculating the frequency component corresponding to the unit section is the same as that conventionally performed.
However, in this case, only one frequency component can be extracted for one unit section. In the present invention, in order to extract a plurality of frequency components from one unit section, the processing of step S14 and thereafter, which will be described later, is performed.

After the main frequency component is calculated in step S13, the DC component is removed from the acoustic signal in the unit section (step S14). Here, the removal of the DC component is not performed uniformly over the entire unit section, but for the portion where the acoustic signal is 0 or more,
The average intensity of the part is subtracted from the signal intensity at each time, and even for the part where the acoustic signal is 0 or less,
The average intensity of the part is subtracted from the signal intensity with the positive and negative signs left unchanged (actually, the average intensity is increased). As a result, a corrected acoustic signal in which a large amplitude portion is suppressed can be obtained. Figure 5
A corrected acoustic signal obtained by correcting the original acoustic signal shown in (c) is shown in FIG. 5 (d).

Then, the signal is 0 with respect to the corrected acoustic signal.
A sub-zero crossing point that intersects the level is detected (step S
15). This is performed by exactly the same process as the detection of the main zero crossing point in the original acoustic signal in step S11. However, as shown in FIG. 5D, in the corrected acoustic signal, the number of zero-crossing points is obviously increased as compared with the original acoustic signal, so many sub-zero-crossing points are detected. The sub-zero crossing point detected from the corrected acoustic signal shown in FIG. 5 (d) is shown in FIG. 5 (e).

Subsequently, the sub-frequency component is calculated according to the corrected acoustic signal and the detected sub-zero crossing point (step S16). Among the sub-frequency components, the period is given as an average period determined by the sub-zero crossing point t3 which is the start point and the sub-zero crossing point t4 which is the end point in the unit section. In addition, as the amplitude of the sub-frequency component, a value that maximizes the absolute value of the corrected acoustic signal in the unit section is set.

The processes of steps S14 to S16 are repeated a predetermined number of times, and the sub frequency components for the repeated number of times are extracted. As described above, the intensities corresponding to a plurality of frequencies can be detected for each unit section, and by processing over the entire section of the original acoustic signal, a set of frequencies and intensities is set for all the set unit sections. Can be calculated.

When the start time is the start time and the end time is the end time of each unit section, the start time / end time
A frequency / intensity pair will be obtained. This set of start time / end time / frequency / intensity is referred to as phoneme data in this specification. In step S1, a process of deleting phoneme data whose intensity does not reach the predetermined value is also performed. As a result, the state of the remaining phoneme data is shown in FIG.

In FIG. 6, the phoneme data is indicated by a downward triangle with the frequency as the pitch of the sound and the strength as the strength of the sound. The pitch of the sound is represented by the position of the upper side of the triangle in the vertical direction, and the strength of the sound is represented by the height of the triangle. Further, the length of the unit section is represented by the length of the upper side of the triangle. It is preferable that the number of phoneme data selected in the frequency analysis of step S1 is about 16 for each unit section, but in order to avoid making the drawing complicated, in the example of FIG. 6, three unit phonemes are shown for each unit section. There is.

When the phoneme data is selected for each unit section, a block phoneme is created based on the selected phoneme data (step S2). The block phoneme creation processing is performed by connecting phoneme data that are close to each other on the time axis and have similar note numbers and velocities. Then, the average value of each constituent phoneme data is given as the note number of the block phoneme, and the maximum value of each constituent phoneme data is given as the velocity. For each unit section, it is usual that a plurality of about 16 pieces of phoneme data are extracted by frequency analysis, but in that case, a plurality of block phonemes that overlap on the time axis are created. However, since the attribute information in the subsequent stage becomes complicated, it is possible to integrate the block phonemes that temporally overlap with each other by representing only the block phoneme having the largest velocity. The specific method will be described below in Japanese Patent Application No. 2000-351775. It is performed by calculating the unified frequency Nr (t) and the unified strength Vr (t) for each unit section by the following [Formula 3].

[Formula 3] Nr (t) = [Σ {n × Vn (t)}] / ΣVn (t) Vr (t) = [Σ {Vn (t)} ² ] ^1/2

In the above [Formula 3], n is a note number, and Vn (t) is a velocity corresponding to the note number n. Also, the summation calculated by Σ is performed for phoneme data. That is, when 16 pieces of phoneme data are selected, 16 sums are calculated. As the unified frequency Nr (t) by [Equation 3], the frequency at the position of the center of gravity when each phoneme data is plotted on the graph having the frequency axis and the intensity axis is calculated, and the unified strength Vr (t) is obtained. As a result, the mean square value of the intensity distribution is calculated. In this way, a block phoneme having a uniform frequency and a uniform intensity is obtained for each unit section. Further, when the unified frequencies are similar between the adjacent block phonemes, they are integrated over a plurality of unit intervals, and a new block phoneme is obtained. The case where the unified frequencies are similar is the case where the difference in frequency is included in the preset range. As the frequency and intensity of the block phoneme after integration, the average value of the frequency and intensity of the block phoneme before integration is given. FIG. 7 shows the state of block phonemes obtained by the process of step S2 for the phoneme data shown in FIG. In the example of FIG.
The pitch and strength of the sound are ignored and expressed.

Apart from the processing of steps S1 and S2,
A basic cycle is calculated for the input acoustic signal (step S3). This is done by performing an autocorrelation analysis. Specifically, the given acoustic signal g
Relative to (t), it follows by changing the period T _n by [Equation 4] Request period T _n correlation value R (T _n) is maximized.

[Formula 4] R (T _n ) = Σ _t g (t) × g (t + T _n ).

As the cycle T _n , each note number (n
= 0 to 127), 128 patterns are given. FIG. 8 is a diagram showing the relationship between the input acoustic signal g (t) and the acoustic signal g (t + T _n ) shifted by the period T _n . Figure 8 (a)
Shows the input acoustic signal g (t), FIG. 8 (b) shows the acoustic signal g (t + T ₁₂₇ ), and FIG. 8 (c) shows the acoustic signal g (t +).
T _{1 26),} FIG. 8 (d) shows the audio signal g (t + T _n). Among such audio signal g (t + T _n), regarded as the period of the correlation value R (T _n) enter the T _n at which is the maximum acoustic signal g (t) of the original audio signal g (t).

Then, the block phonemes obtained in step S2 are classified based on the cycle T _n obtained in step S3 (step S4). Specifically, the periods T _n
Compare the note numbers of block phonemes that are only apart. When the difference between the note numbers falls within the set predetermined value range, both blocks are classified into the same channel. The number of channels to be classified differs depending on the number of block phonemes and the range of difference between note numbers that are regarded as the same,
Divide into more channels than the number of channels of the final output data. For example, the block data shown in FIG. 7 is classified into eight channels as shown in FIG.

Next, a heart sound attribute is added to the classified block phonemes of each channel (step S5). Specifically, a database regarding heart sounds was prepared in advance, and the block phoneme patterns stored in this database were compared with the block phoneme patterns of each classified channel, and it was determined that they were similar. In case,
The heart sound attribute registered in the database is added to the channel. The block phoneme pattern is an array pattern of phoneme data on which each block phoneme is created. To each block phoneme, specifically, three types of attribute information, I sound, II sound, and others are given.
This determination is made based on the following four rules.

Pitch: The II note has a higher pitch than the I note. (Sound II> I
Sound length: Sound length is fixed to short for both I and II sounds, and II sound is shorter than I sound. (I sound and II sound <length threshold, and I sound> II sound) Sound intensity: Both I sound and II sound are stronger than other components,
The balance of the I and II sounds changes depending on the position of the sensor to be heard. As an exception, I and II sounds such as reflux murmur
May be stronger than sound. (I sound and II sound> strength threshold) Sound interval: The interval between I sound and II sound is shorter than the interval between II sound and I sound. As the heart rate increases, the interval between sounds II and I becomes shorter, but the interval between sounds I and II does not change, that is, approaches the interval between sounds I and II. (I-II interval <II-I interval)

As a result, the block phoneme shown in FIG. 9 is provided with the attribute information as shown in FIG. 10 for each channel. Subsequently, the attribute information corresponding to each channel is added with reference to the database recording more detailed information. Based on the phoneme data array pattern that is the basis for creating a block phoneme, first, for the block phonemes to which the attribute information of the I and II sounds is added, the detailed classification of the block phonemes For phonemes, refer to the database in which the detailed attribute information is registered for the specific attribute information, based on the phoneme data array pattern, and sound III, IV (or III and IV combined sound), heart murmur (contraction) Phase, diastole, continuity, subflow classification such as regurgitation or driveability), click sound, fricative sound, etc. are determined, and corresponding detailed attribute information is added.
In particular, it is possible to specify whether the sound I is the mitral valve component, the tricuspid valve component, or the aortic valve component, and the sound II is the aortic valve component or the pulmonary valve component. . FIG. 11 shows a block phoneme to which detailed attribute information is added for each channel.

In step S5, it is possible to specify what kind of heart sound attribute the data of each channel has, but it is also possible to specify it by a person, not automatically. In this case, a state in which the block phonemes are classified into a plurality of channels as shown in FIG. 9 is displayed on the screen or the like. In the example of FIG. 9, the block state is shown, but actually, the state in which the phoneme data as shown in FIG. 6 is superimposed is displayed. By viewing such a display, it is possible to confirm both the state and the period of the phoneme data forming each block phoneme at the same time.

A person such as a doctor who can judge the heart sound attribute by the heart sound waveform assigns the heart sound attribute to each channel while observing the frequency element. As shown in FIG. 9, unlike the conventional electrocardiographic waveform, even block phonemes appearing at a predetermined cycle interval appear in the same channel, so that more accurate determination is possible. For example, when only one block phoneme is viewed, even if it is likely to be determined to be a heart sound component, a component that does not appear periodically can be determined to be noise and can be excluded.

Subsequently, the block phoneme to which the attribute information is added is further structured into a document (step S6). FIG. 12 shows an example of the source code when the XML (eXtensible Markup Language) standard is adopted as the structured document. The example shown in FIG. 12 is a structural document of one cycle of the heart sound from the head of the block phoneme shown in FIG. 11, and is described in 37 lines. The block phonemes described in FIG. 12 are described in compliance with MIDI. XML
Is basically a format in which actual data is surrounded by a pair of tags. In FIG. 12, the start and end of a document are defined by a pair of tags on the first line at the beginning and the 37th line at the end. In the pair of tags described in the second line, the pronunciation start time (<StartTime>) defined in the MIDI standard-compliant event data (defined by the <Event> tag) described from the start to the end of this document. The unit of time (resolution per second) of the pronunciation end time (defined by <EndTime> tag) is defined. 3rd line <H
The eartCycle> tag and the end tag on the 36th line </ HeartCycle> indicate that the data for one cycle of the heart sound is described below. 4th to 14th lines: Heart sound No. I <FirstSou
nd> is described. The 4th and 14th lines are the heart sounds I
It is a pair of tags indicating that it is a sound. This heart sound No. 1 sound is further classified into three detailed attributes. The 5th to 7th lines show the mitral valve component M1, the 8th to 10th lines show the tricuspid valve component T1, and the 11th to 13th lines show the aortic valve component, respectively. Each classification has the same configuration. For example, in the case of the mitral valve component M1, the sixth row and the seventh row
Each of the lines describes the unified code data according to the MIDI standard. For example, <StartTime on line 6
> And <EndTime> describe a time expressed as a relative time called a delta time in the MIDI standard as an absolute time, and indicate that a sound is produced from time “10” to time “20”. Also, <Pitch> on the 7th line corresponds to the MIDI standard note number, and the note number “3
It indicates that the sound is produced at the pitch corresponding to "2". 7
The <Level> on the line corresponds to the velocity of the MIDI standard, and indicates that the tone is produced with the strength of the tone corresponding to the velocity “60”.

When the created structured document is in the XML format, it can be browsed by the Internet browser and the heart sound can be reproduced as an acoustic signal by using the dedicated player software and the MIDI sound source. Specifically, the created structured document is registered in the WWW server, and the user activates the browser on his or her personal computer and accesses the WWW server on the Internet to obtain the XML document. Next, the MID plugged into the browser
The I sequencer software is the M that is recorded in the XML document.
The acoustic signal is reproduced while controlling the MIDI sound source according to the IDI data. In this way, the heart sounds are encoded and X
Recording in ML format is not only convenient for distribution via the Internet and suitable for reproduction using a MIDI sound source, but in recent years it has been digitized as an electronic medical record using XML format. It is also effective from the viewpoint of consistency with existing medical record information.

When the MIDI data to which the heart sound attribute is added is stored as an electronic medical record, in addition to the above-described effect that the heart sound can be reproduced by the MIDI sound source, the quantitative pathological condition necessary for diagnosis can be read from the document as described below. There is an effect that can be. For example, sound I and sound II have pathological conditions such as acceleration, attenuation, and division, and these pathological conditions can be judged from the numerical values of MIDI event data surrounded by attributes. That is, if the velocity (strength) enclosed by the <Level> tag or the note number (height) enclosed by the <Pitch> tag is higher than a predetermined value, it is an increase, and if it is lower than the predetermined value, it is an attenuation. In the example of sounds I and II in FIG. 12, two events (notes represented by triangles) that are close to each other on the time axis are configured, but if they are temporally separated, it can be determined that they are split. Ie <E of the previous event
If the difference between the value of the ndTime> tag and the value of the <StartTime> tag of the subsequent event is equal to or greater than the predetermined value, it can be determined that the split. Also, in the clinical diagnosis of heart murmurs, there are often Levine classifications (classes 1 to 6; 1 is a weak noise that is difficult to hear by auscultation and cannot be distinguished unless it is an phonocardiogram, and 6 is without a stethoscope). Strong noise that can be heard just by being on the side is used, but this can also be identified from the velocity (strength) enclosed by the <Level> tags.

Next, an apparatus for realizing the above-described time-series signal analysis method according to the present invention will be described. FIG.
FIG. 3 is a functional block diagram showing an embodiment of a time series signal analysis device. In FIG. 13, 1 is a time-series signal acquisition means, 2 is a frequency analysis means, 3 is a block phoneme creation means,
Reference numeral 4 is a basic period calculation means, 5 is a block phoneme classification means, and 6
Is an attribute information giving unit, 7 is an attribute database, 8 is a detailed attribute database, 9 is an output unit, and 10 is a structured document creating unit. The time-series signal analysis device shown in FIG. 13 is actually realized by connecting audio-related peripheral devices to a computer and installing a dedicated program.

The time-series signal acquisition means 1 is for acquiring time-series signals. When acquiring a heart sound signal, a microphone is attached to a stethoscope and the acquired heart sound is PCM.
Etc. are realized by a sampling device that converts into a digital signal. The frequency analysis unit 2 has a function of performing frequency analysis of the acquired time series signal to create phoneme data, and specifically, executes the process of step S1 of FIG. 2 according to a computer program. The block phoneme creating unit 3 has a function of creating a block phoneme based on the phoneme data obtained by the frequency analyzing unit 2, and specifically executes the process of step S2 of FIG. 2 in accordance with a computer program. The basic cycle calculating means 4 has a function of performing autocorrelation analysis of a time-series signal to calculate a basic cycle. Specifically, the basic cycle calculating means 4 executes the process of step S3 of FIG. 2 according to a computer program. The block phoneme classifying unit 5 has a function of classifying the block phonemes into a plurality of groups based on the block phoneme created by the block phoneme creating unit 3 and the basic cycle calculated by the basic cycle calculating unit 4, First, the process of step S4 in FIG. 2 is executed according to the computer program.
The attribute information assigning means 6 assigns the attribute information of the heart sound to each channel into which the block phonemes are classified according to a preset rule, and sets the phoneme data array pattern which is the basis for creating the block phonemes. Based on this, it has a function of further adding detailed attribute information extracted from the detailed attribute database 8. Specifically, the process of step S5 of FIG. 2 is executed according to the computer program. The attribute database 7 is
A detailed attribute database 8 is a database in which correspondences between phoneme data array patterns and respective heart sound attributes are recorded in order to classify sounds, second sounds, and others. The detailed attribute database 8 specifies the first and second sounds in more detail. Therefore, for sound I, record the array pattern of phoneme data corresponding to each of the mitral valve component, tricuspid valve component, and aortic valve component, and for sound II, record the aortic valve component and pulmonary valve component, respectively. The sequence pattern of the phoneme data corresponding to is recorded, and the heart sounds classified as “other” are recorded as the third and fourth sounds (or I
II and IV), heart murmur (systolic or diastolic or continuous, subclasses such as regurgitant or driveable are also defined), clicks, fricatives Each phoneme data array pattern is recorded.
The output unit 9 has a function of displaying / printing the classified block phonemes. Here, the output block phoneme is not useful for automatically assigning the heart sound attribute in step S5 of FIG. 2 described above, but is useful for medical personnel to assign the heart sound attribute. By looking at the block phonemes classified for each channel, it becomes possible to assign the heart sound attribute in consideration of the fundamental period. Specifically, the output unit 9 is realized by a display device such as a CRT or a liquid crystal display, or a printer device of various monochrome or color printing systems. Structured document creating means 10
Is a block phoneme with heart sound attribute information added to XML
It has a function of converting it into text information such as a standard, and specifically executes the process of step S6 of FIG. 2 in accordance with a computer program.

Although the preferred embodiments of the present invention have been described above, the present invention is not limited to the above embodiments and various modifications can be made. In the above embodiment, the case of analyzing the heart sound as the time-series signal has been described, but it is also applicable to the analysis of other bioacoustic signals such as respiratory sounds, non-acoustic bio-time-series signals such as electrocardiogram, and in other fields, for example, engine It is also possible to analyze an acoustic signal such as sound as a time-series signal and apply it to a failure diagnosis of an automobile engine.

[0051]

As described above, according to the present invention,
By performing frequency analysis on a time-series signal for each given unit interval set for a given time-series signal, a plurality of phoneme data composed of a set of frequency and intensity for each unit interval Is created, out of the created plurality of phoneme data, one that satisfies a predetermined intensity is extracted,
Based on the similarity of the start time, end time, and frequency of the extracted phoneme data, create a block phoneme consisting of parameters of start time, end time, frequency, and intensity, By performing a correlation analysis, the basic period of this time-series signal is calculated, and if the frequencies are similar among the plurality of block phonemes and the start time difference is close to an integral multiple of the calculated basic period, The effect that the medical staff such as a doctor can easily judge the heart sound attribute because the block phonemes are displayed in the same group and the classified group units are displayed. Play. Furthermore, by using the knowledge database in which the heart sound waveform is registered in association with the heart sound attribute including the characteristics and cycle of each block, it is possible to automatically add the heart sound attribute.

[Brief description of drawings]

FIG. 1 is a diagram showing a basic principle of signal analysis in a time-series signal analysis device of the present invention.

FIG. 2 is a flowchart showing an outline of analysis of a time series signal according to the present invention.

FIG. 3 is a diagram showing an example of a heart sound waveform captured as a time-series signal.

FIG. 4 is a flowchart showing details of step S1 in FIG.

FIG. 5: Setting and cycle of unit section in acoustic signal
It is a figure for demonstrating determination of amplitude.

6 is a diagram showing phoneme data generated by frequency analysis from the heart sound waveform shown in FIG. 3;

7 is a diagram showing block phonemes calculated based on the phoneme data shown in FIG.

FIG. 8 is a diagram showing a relationship between an input acoustic signal g (t) and an acoustic signal g (t + T _n ) shifted by a period T _n .

9 is a diagram showing a state in which the block phonemes shown in FIG. 7 are classified based on a fundamental cycle.

10 is a diagram showing a state in which attribute information is added to each block phoneme shown in FIG.

11 is a diagram showing a state in which more detailed attribute information is added to each block phoneme shown in FIG.

FIG. 12 is a diagram showing a source code when the XML standard is adopted as a structured document.

FIG. 13 is a functional block diagram showing an embodiment of a time-series signal analysis device according to the present invention.

[Explanation of symbols]

1. Time-series signal acquisition means 2 ... Frequency analysis means 3 ... Block phoneme creating means 4 ... Basic period calculation means 5 ... Block phoneme classifying means 6 ... Attribute information adding means 7 ... Attribute database 8 ... Detailed attribute database 9 ... Output means 10: structured document creation means

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁷ Identification code FI theme code (reference) // A61B 5/025 A61B 5/02 350

Claims (6)

[Claims] 1. A frequency analysis means for generating a plurality of phoneme data composed of a set of frequency and intensity for each predetermined unit section set for a given time-series signal, and the plurality of generated plurality of phoneme data. From the phoneme data, the ones that satisfy the specified strength are extracted, and the start time / end time / frequency is based on the similarity of the start time / end time / frequency of the extracted phoneme data.
Block phoneme creating means for creating a block phoneme composed of parameters of end time, frequency, and intensity, and a basic cycle for calculating a basic cycle of the time series signal by performing autocorrelation analysis on the time series signal. Frequency is similar to the calculation means, the plurality of block phonemes,
And a block phoneme classifying unit that classifies the difference in start time close to an integral multiple of the calculated basic period into the same group, and a display unit that displays block phonemes in the classified group units. A time-series signal analysis device comprising: 2. A frequency analysis means for generating a plurality of phoneme data composed of a set of frequency and intensity for each predetermined unit section set for a given time series signal, and the plurality of generated plurality of phoneme data. From the phoneme data, the ones that satisfy the specified strength are extracted, and the start time / end time / frequency is based on the similarity of the start time / end time / frequency of the extracted phoneme data.
Block phoneme creating means for creating a block phoneme composed of parameters of end time, frequency, and intensity, and a basic cycle for calculating a basic cycle of the time series signal by performing autocorrelation analysis on the time series signal. Frequency is similar to the calculation means, the plurality of block phonemes,
And a block phoneme classifying unit that classifies the difference in start time close to an integral multiple of the calculated basic period into the same group, and for each of the grouped phoneme blocks, in a predetermined database. A time-series signal analysis device comprising: an attribute information assigning unit that assigns the defined attribute information; and an output unit that outputs the block phoneme to which the attribute information is assigned. 3. Frequency analysis means for generating a plurality of phoneme data composed of a set of frequency and intensity for each predetermined unit section set for a given time series signal, and the plurality of generated plurality of phoneme data. From the phoneme data, the ones that satisfy the specified strength are extracted, and the start time / end time / frequency is based on the similarity of the start time / end time / frequency of the extracted phoneme data.
Block phoneme creating means for creating a block phoneme composed of parameters of end time, frequency, and intensity, and a basic cycle for calculating a basic cycle of the time series signal by performing autocorrelation analysis on the time series signal. Frequency is similar to the calculation means, the plurality of block phonemes,
And the difference in start time close to an integral multiple of the calculated basic period, a block phoneme classifying unit for classifying into a same group, for each of the grouped block phonemes, in a predetermined database A time series characterized by comprising: attribute information giving means for giving the defined attribute information; and structured document creating means for outputting the block phoneme to which the attribute information is given as hierarchically structured text information. Signal analyzer. 4. The attribute information assigning means preferentially assigns a block phoneme that matches the database, and a block phoneme for which there is no attribute information to be assigned to the database belongs to the same group. The time-series signal analysis device according to claim 2 or 3, wherein attribute information of a block phoneme to which attribute information is added is added. 5. The time-series signal is a heart sound signal, the basic cycle is a heartbeat cycle, and the attribute information is M1, T1, A1 and second sound constituent components of the I-th sound component of the heart sound. The time series signal analysis apparatus according to any one of claims 2 to 4, wherein the database is A2, P2, etc., and the database associates attribute information with block phoneme patterns. 6. The text information output by the structured document creating means is described in conformity with the XML standard format,
4. The time series according to claim 3, wherein the included block phonemes are described in a MIDI format having delta time information corresponding to a section, note number information corresponding to a frequency, and velocity information corresponding to an intensity. Signal analyzer.