JP5092876B2 - Sound processing apparatus and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce an arithmetic amount required for specifying a beat of a sound signal. <P>SOLUTION: A window sequence setting section 72 sets a window sequence GA in which a plurality of search windows WA are arranged from a candidate point BC with an interval responding to a beat cycle T of the sound signal S, by making each of a plurality of peaks PB in a time sequence of a beat index OB for indicating a change amount of intensity of the sound signal S, as the candidate point BC of the beat. A selection section 74 selects the candidate point BC corresponding to n2 pieces of window sequences GA in which the number of search windows WA including peak PB in the plurality of window sequences GA are large. A window sequence setting section 76 sets the window sequence GB in which a plurality of search windows WB are arranged from the candidate point BC with the interval corresponding to the beat cycle T, for each of the n2 pieces of candidate points. A beat determination section 80 determines the candidate point BC as the beat B, when a confidence degree R in which the beat index OB in the peak PB in each search window WB of the window sequence GB is averaged, becomes largest. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a technique for detecting a beat point from an acoustic signal representing an acoustic (speech or musical sound) waveform.

Various techniques for detecting beat points from acoustic signals have been proposed. For example, Patent Document 1 discloses an onset curve indicating an amount of change in intensity of an acoustic signal over time and a plurality of beat pattern templates (beats) arranged on the time axis at intervals according to the tempo of the music. and a technique for detecting beat points of an acoustic signal using a pattern template). Confidence that indicates whether each peak of the onset curve corresponds to a beat point by comparing the onset curve and the beat pattern template for each of a plurality of frames defined on the time axis level) is calculated. The beat point of the acoustic signal is determined according to the level of reliability.
US Pat. No. 7,183,479

  However, in the technique of Patent Document 1, the reliability is calculated for every frame from the start point to the end point of the acoustic signal. Therefore, there is a problem that enormous calculation is required for specifying beat points. In view of the above circumstances, an object of the present invention is to reduce the amount of calculation required for specifying the beat point of an acoustic signal.

  In order to solve the above problems, the acoustic processing device according to the first aspect of the present invention provides a beat index that specifies a time series (for example, a beat index series X) of beat index values indicating a change in a feature amount of an acoustic signal. Identifying means; beat period identifying means for identifying a beat period of an acoustic signal; candidate detecting means for detecting a plurality of peaks in a time series of beat index values; and each of the plurality of peaks as beat point candidate points. First window row setting means for setting a first window row (for example, window row GA in FIG. 10) in which a plurality of search windows are arranged from the candidate points at intervals according to the period, and beats of the plurality of first window rows. A selection means for selecting candidate points corresponding to two or more window rows having a large number of search windows including a peak of the index value, and of the two or more candidate points selected by the selection means, Selected according to the beat index value of the peak in each search window arranged at the corresponding interval Beat point determination means for determining a candidate point as a beat point.

  In the above configuration, from the candidate points corresponding to two or more window rows selected according to the number of search windows including the time-series peak of the beat index value, the candidate point corresponds to the beat cycle. Candidate points corresponding to peak beat index values in each search window arranged at intervals are specified as beat points. Therefore, the amount of calculation required for specifying the beat point is smaller than that in the configuration in which the beat index value in each search window is examined for all the first window rows set by the window row setting means and the beat point is determined. There is an advantage that it is reduced.

  A preferred aspect of the sound processing apparatus according to the first aspect is based on the window width of each search window in the first window row (for example, the window width w1 in FIG. 10) for each of the two or more candidate points selected by the selection means. A second window array (for example, window array GB in FIG. 10) in which a plurality of search windows having a narrow window width (for example, window width w2 in FIG. 10) are arranged from the candidate points at intervals corresponding to the beat period is set. A window sequence setting unit is provided, and the beat point determination unit determines a candidate point selected according to the peak beat index value in each search window among the plurality of second window columns as a beat point. In the above aspect, since the window width of each search window of the second window row used for selection of candidate points is narrower than the window width of each search window of the first window row, factors other than the beat point of the acoustic signal The possibility that the peak of the beat index value due to (for example, noise) is included in the search window of the second window row is reduced. Therefore, for example, a configuration in which the first window row is used when beat points are determined from two or more candidate points selected by the selection means (that is, beats of peaks in each search window of the first window row corresponding to the candidate points). Compared with a configuration in which the candidate point selected according to the index value is determined as a beat point, it is possible to accurately specify the beat point. On the other hand, when the window width of each search window in the first window row is narrowed, there is a problem that a beat point search is likely to be omitted. In the above aspect, since the window width of each search window of the first window row is wider than the window width of each search window of the first window row, there is an advantage that omission of searching for beat points can be suppressed.

  In a preferred aspect of the sound processing apparatus according to the first aspect, the selecting means includes a search including a peak of a beat index value among a first number (for example, H1 in FIG. 10) of search windows in the first window row. First extraction means for selecting a plurality of first window rows having a large number of windows from a plurality of first window rows set by the first window row setting means; and a first number exceeding the first number in the first window row. A second extraction in which two or more first window rows having a large number of search windows including the peak of the beat index value among the two number of search windows are selected from the two or more first window rows detected by the first extraction means. Means. In the above aspect, the narrowing of the first window row according to the number of search windows including the peak among the first number of search windows of each first window row, and the second number of search windows exceeding the first number. The first window row is narrowed down step by step according to the number of search windows including peaks. Therefore, there is an advantage that the first window row corresponding to the beat point can be accurately extracted as compared with the configuration in which the first window row is selected only from the number of peaks in each first number of search windows. In addition, since the number of objects to be processed by the second extraction unit is reduced in advance, the processing amount of the selection unit is smaller than the configuration in which the first window row is selected only from the number of peaks in each of the second number of search windows. There is an advantage of being reduced.

The sound processing apparatus according to the second aspect of the present invention includes a beat index specifying means for specifying a time series of beat index values indicating a change in a characteristic amount of an acoustic signal, and a beat cycle specifying means for specifying a beat cycle of the sound signal. And candidate detecting means for detecting a plurality of peaks in the time series of beat index values, and each of the plurality of peaks as beat point candidate points, and a plurality of search windows arranged from the candidate points at intervals according to the beat period First window row setting means for setting the first window row, and two or more first windows having a large number of search windows including the peak of the beat index value among the first number of search windows in the first window row. First extraction means for selecting a row from a plurality of first window rows set by the first window row setting means, and a beat index value of a second number of search windows exceeding the first number in the first window row. the first window sequence number encompassing search window peaks is maximized, 2 or more of the first extracting means detects Sorted from 1 window column, and second extracting means for determining the candidate points corresponding to the first window column as beat positions. In the above aspect, the narrowing of the first window row according to the number of search windows including the peak among the first number of search windows of each first window row, and the second number of search windows exceeding the first number. The first window row is narrowed down according to the number of search windows including the peak. Therefore, there is an advantage that the first window row corresponding to the beat point can be accurately extracted as compared with the configuration in which the first window row is selected only from the number of peaks in each first number of search windows. In addition, since the number of objects to be processed by the second extraction unit is reduced in advance, the processing amount of the selection unit is smaller than the configuration in which the first window row is selected only from the number of peaks in each of the second number of search windows. There is an advantage that the amount of calculation necessary for specifying the beat point is reduced as a whole.

  In the aspect including the first extraction means, the first number of search windows is a plurality of search windows selected in order from the start side of the acoustic signal among the plurality of search windows of the first window row. Considering the tendency that beat points often appear clearly and stably in the first part of the music, according to the above-described aspect, the first number of search windows are sequentially selected from the start point side of the acoustic signal. There is an advantage that beat points can be detected accurately and stably compared to the case of using a partial search window.

  In the specific example of the sound processing apparatus according to each of the above aspects (including both the first aspect and the second aspect), the first window row setting unit is configured to perform only the beat period from the peak included in one search window. The next search window is set with the elapsed time as the center of the window width (for example, part (B) and part (C) in FIG. 12). In the above specific example, since the peak in the search window is used as a starting point for setting the next search window, each beat point can be accurately detected even when the beat cycle of the acoustic signal is fluctuated. There are advantages. In the aspect including the second window row setting means, the second window row setting means sets the next search window at the center of the window width at the time when the beat period has elapsed from the peak included in one search window. A configuration for setting is also adopted.

  In the specific example of the sound processing apparatus according to each aspect described above, the beat point determination unit detects, as beat points, peaks in each search window arranged at intervals according to the beat period from candidate points determined as beat points. . In the above specific example, it is possible to detect a time series of beat points over the entire acoustic signal. Further, according to the configuration including the beat point supplementing means for setting a new beat point between the beat points when the beat point interval determined by the beat point determining means exceeds the beat cycle, the acoustic signal There is an advantage that a beat point with a relatively small change in feature value (that is, a beat point that does not appear as a significant peak in the time series of beat index values) can be appropriately identified.

  In the specific examples of the sound processing device according to each of the above aspects, the beat period specifying means calculates the correlation calculation means for calculating the autocorrelation value for each unit section into which the time series of the beat index values is divided, and the correlation calculation means Average means for averaging the autocorrelation values for a plurality of unit intervals, and the beat period is specified from the autocorrelation values after averaging by the averaging means. In the above specific examples, since the average of autocorrelation values over a plurality of unit intervals is used for specifying the beat cycle, for example, beats are calculated only from the autocorrelation values of a specific interval (for example, one unit interval) of an acoustic signal. Compared with the case where the period is specified, there is an advantage that the beat period can be specified with high accuracy by reducing the influence of the peak of the beat index value caused by factors other than the beat point of the acoustic signal (for example, noise).

  In the specific examples of the acoustic processing device according to each of the above aspects, the beat index specifying means includes an index calculation means for calculating a time series of beat index values indicating changes in the feature amount of the acoustic signal, and a beat calculated by the index calculation means. Peak emphasizing means for emphasizing increase / decrease of beat index values in the time series of index values. In the above specific example, since the increase / decrease (peak) of the beat index value in the time series of the beat index value is emphasized, the advantage that the beat point can be accurately identified as compared with the case where the increase / decrease of the beat index value is not emphasized. There is.

The sound processing apparatus according to all the aspects described above is realized by hardware (electronic circuit) such as a DSP (Digital Signal Processor) dedicated to processing of input sound, and a general purpose such as a CPU (Central Processing Unit). This is also realized by cooperation between the arithmetic processing unit and the program .
The program according to the first aspect of the present invention includes a beat index specifying process for specifying a time series of beat index values indicating a change in a characteristic amount of an acoustic signal, a beat cycle specifying process for specifying a beat period of the acoustic signal, Candidate detection processing for detecting a plurality of peaks in the time series of beat index values, and a plurality of search windows arranged from the candidate points at intervals according to the beat period, with each of the plurality of peaks as a candidate point of a beat point First window sequence setting processing for setting one window sequence, and selection for selecting candidate points corresponding to two or more window sequences having a large number of search windows including a peak of the beat index value among a plurality of first window sequences Among the two or more candidate points selected in the process and the selection process, the candidate point selected according to the beat index value of the peak in each search window arranged at intervals according to the beat period from the candidate point as a beat point Causes the computer to execute beat point determination processing to be determined. According to the program of this invention, the effect | action and effect similar to the sound processing apparatus which concern on a 1st aspect are implement | achieved.
The program according to the second aspect of the present invention includes a beat index specifying process for specifying a time series of beat index values indicating a change in a characteristic amount of an acoustic signal, and a beat cycle specifying process for specifying a beat period of the acoustic signal. , Candidate detection processing for detecting a plurality of peaks in the time series of the beat index value, and using each of the plurality of peaks as a candidate point for a beat point, a plurality of search windows at intervals according to the beat period A first window row setting process for setting the first window row arranged from the above, and a number of search windows including a peak of a beat index value among a first number of search windows in the first window row is greater than 2 A first extraction process for selecting a first window row from a plurality of first window rows set in the first window row setting process; and a second number of search windows exceeding the first number in the first window row. The first window row in which the number of search windows including the peak of the beat index value is the maximum, Serial The sorted from two or more first window sequence detected by the first extraction process, to execute a second extraction process to determine the candidate points corresponding to the first window column as beats the computer. According to the program of this invention, the effect | action and effect similar to the sound processing apparatus which concern on a 2nd aspect are implement | achieved.
The program according to each aspect of the present invention is provided to the user in a form stored in a computer-readable recording medium and installed in the computer, and is also provided from the server device in the form of distribution via a communication network. Installed on the computer.

  FIG. 1 is a block diagram of a sound processing apparatus according to an embodiment of the present invention. As shown in FIG. 1, the sound processing device 100 is realized by a computer system including a control device 12 and a storage device 14. The control device 12 is an arithmetic processing device that functions as a plurality of elements (frequency analysis unit 20, beat index specification unit 30, beat cycle specification unit 40, candidate detection unit 60, beat point specification unit 70) by executing a program ( CPU). However, each element of the control device 12 is also realized by a dedicated electronic circuit (DSP). The storage device 14 stores a program executed by the control device 12 and various data used by the control device 12. A known recording medium such as a semiconductor storage device or a magnetic storage device is arbitrarily adopted as the storage device 14.

  The storage device 14 stores an acoustic signal S for each piece of music. The acoustic signal S is a time series of a large number of samples obtained by sampling a waveform on the time axis of a musical performance sound (musical sound or voice) at a sampling frequency of 44.1 kHz, for example. The control device 12 specifies the beat period (tempo) T and each beat point (beat position) B of the acoustic signal S. The beat period T is a time interval between successive beat points B. In this embodiment, it is assumed that the beat period T is constant throughout the music represented by the acoustic signal S (for example, music whose tempo is substantially constant throughout the music, such as dance music and pop music).

  As shown in FIG. 2, the frequency analysis unit 20 performs a Fourier transform (for example, a short-time Fourier transform using a Hanning window) on each of a plurality of frames F obtained by dividing the acoustic signal S on the time axis. The frequency spectrum (power spectrum) Q of F is specified. Each frame F corresponds to, for example, 2048 samples of the acoustic signal S, and the shift amount (hop size) of the adjacent frames F corresponds to, for example, 200 samples of the acoustic signal S. Therefore, the frames F that follow each other overlap each other on the time axis.

  The beat index specifying unit 30 in FIG. 1 specifies a beat index series X indicating a change over time in the feature amount of the acoustic signal S. The beat index series X of this embodiment has N beat index values OB (OB [1], OB [2], N, which are natural numbers) calculated for each frame F according to the amount of change in the intensity of the acoustic signal S. ……, OB [N]).

As shown in FIG. 3, the beat index specifying unit 30 of this embodiment includes an index calculation unit 32 and a peak enhancement unit 34. The index calculation unit 32 converts the time series of N beat index values OA (OA [1] to OA [N]) according to the amount of change in the intensity of the acoustic signal S in each successive frame F to the beat index series X0. Calculated as More specifically, the j-th (j = 1 to N) beat index value OA [j] in the beat index series X0 is the value of the j-th frame F, as exemplified by the following equation (1). This is a numerical value obtained by summing (or averaging) the difference (absolute value) in intensity at each frequency between the frequency spectrum Q and the frequency spectrum Q of the (j + 1) th frame F. The intensity A [j, f] in the arithmetic expression (1) means the intensity at the frequency f in the frequency spectrum Q of the j-th frame F.

  Part (A) of FIG. 4 is a curve representing the beat index series X0 (that is, a curve connecting beat index values OA [1] to OA [N]). As shown in part (A) of FIG. 4, a peak appears at the time (including the beat point of the music) when the intensity of the acoustic signal S changes greatly in the beat index series X0.

3 emphasizes the peak of the beat index series X0 calculated by the index calculation section 32 (rapid increase / decrease in the beat index value OA) to thereby increase the beat index series X (beat index value OB [1], OB [2], ..., OB [N]) is specified. More specifically, the peak emphasizing unit 34 uses a numerical value MA [j] (hereinafter referred to as “smooth value”) MA [j] obtained by suppressing (smoothing) increase / decrease in the beat index series X0 as the beat index value OA [j] The beat index value OB [j] is calculated by subtracting from (OB [j] = OA [j] −MA [j]). The smooth value MA [j] is an average value (low frequency component) of a plurality of beat index values OA in the vicinity of the beat index value OA [j]. For example, the following formula (2C) to formula (2C) Defined by

  As shown in FIG. 5 and the arithmetic expression (2B), basically, n beat index values OA (OA [j− (n−) with the beat index value OA [j] as the center in the beat index series X0. The average value (moving average) of 1) / 2] to OA [j + (n-1) / 2]) is calculated as the smoothed value MA [j]. On the other hand, as shown in FIG. 5, n beat index values OA centered on the beat index value OA [j] cannot be secured in the vicinity of the start point of the acoustic signal S, so FIG. 5 and the calculation formula (2A) As shown, the average value of (n + 1) / 2 beat index values OA from the beat index value OA [j] to the beat index value OA [j + (n-1) / 2] is the smoothed value MA [ j]. Similarly, the n beat index values OA centered on the beat index value OA [j] cannot be secured in the vicinity of the end point of the acoustic signal S, so that the beat is calculated as shown in FIG. 5 and the arithmetic expression (2C). An average value of (N−j + 1) beat index values OA from the index value OA [j] to the last beat index value OA [N] is calculated as the smoothed value MA [j]. Since the beat index value OB [j] is calculated by subtracting the smooth value MA [j] calculated by the above method from the beat index value OA [j], as shown in part (B) of FIG. The curve of the beat index series X (that is, the curve connecting the beat index values OB [1] to OB [N]) has a shape that emphasizes the peak (high frequency component) of the beat index series X0. As shown in part (B) of FIG. 4, the peak PB appears at the time point (typically the beat point of the music) at which the intensity of the acoustic signal S greatly changes in the beat index series X.

  The beat period specifying unit 40 in FIG. 1 specifies the beat period T of the acoustic signal S. For specifying the beat period T, the beat index series X (OB [1] to OB [N]) specified by the beat index specifying unit 30 is used. FIG. 6 is a block diagram of the beat period specifying unit 40. The section setting unit 42 in FIG. 6 divides the beat index series X into NU (NU is a natural number) unit sections U. As shown in part (C) of FIG. 4, each unit section U is a section corresponding to, for example, 1024 beat index values OB (1024 frames F), and a difference between successive unit sections U. The amount (hop size) corresponds to, for example, 128 beat index values OB (frame F). Therefore, successive unit sections U overlap each other on the time axis.

  The correlation calculation unit 44 in FIG. 6 calculates the autocorrelation value C0 for each unit section U by executing the autocorrelation calculation of the beat index series X for each of the NU unit sections U. In the autocorrelation calculation, for each of a plurality of time differences L sequentially changed in units of one beat index value OB, the beat index series X to which the time difference L is given and the original beat index series X (that is, the time difference L Is a process of calculating the autocorrelation value C0 of the beat index value OB with the beat index series X) to which is not added. Part (A) of FIG. 7 is a graph showing the relationship between the time difference L and the autocorrelation value C0. The autocorrelation value C0 increases at the time difference L that matches the interval between the peaks PB of the beat index series X.

  By the way, the beat index value OB used for calculating the autocorrelation value C0 in each unit section U (that is, the beat index in the section where the beat index series X given the time difference L and the original beat index series X overlap). The number m of the values OB) decreases as the time difference L increases. Therefore, as shown in part (A) of FIG. 7, as the time difference L increases, the fluctuation of the autocorrelation value C0 decreases and becomes more susceptible to noise. Therefore, the correlation calculation unit 44 of this embodiment calculates the autocorrelation value C by normalizing the autocorrelation value C0 using the number m of beat index values OB used for calculating the autocorrelation value C0. The number m is a numerical value obtained by subtracting the total number mL of beat index values OB within the time difference L from the total number mU of beat index values OB included in the unit interval U (1024 in this embodiment) (m = mU−mL). The correlation calculation unit 44 calculates the autocorrelation value C by dividing the autocorrelation value C0 by the number m (C = C0 / m). By normalizing the autocorrelation value C0 as described above, an appropriate autocorrelation value C can be calculated even when the time difference L is large, as shown in part (B) of FIG. As shown in part (B) of FIG. 7, a peak appears in the autocorrelation value C at a time difference L corresponding to an integral multiple of the interval between peaks PB of the beat index series X (that is, the interval between beat points).

  As described above, the relationship between the time difference L and the autocorrelation value C is specified for each of the NU unit intervals U. Part (A) of FIG. 8 is a conceptual diagram in which the time difference L (horizontal axis) is shared and the autocorrelation values C of the NU unit sections U are arranged on the vertical axis (number of the unit section U). The solid line extending in the vertical direction in the part (A) of FIG. 8 indicates the peak apex of the autocorrelation value C of each unit section U. In this embodiment, since it is assumed that the beat period T (tempo) is fixed over the entire acoustic signal S, the peak of the autocorrelation value C in each unit section U is a time difference as shown in part (A) of FIG. It arrange | positions at substantially equal intervals along an axis | shaft.

  The average unit 46 in FIG. 6 calculates the average correlation value Cave by averaging (or summing) the autocorrelation values C calculated for each of the NU unit intervals U by the correlation calculation unit 44 for each time difference L. That is, the average correlation value Cave corresponding to a specific time difference L is obtained by averaging the autocorrelation values C at the time difference L over NU unit intervals U as shown in part (A) and part (B) of FIG. It is a numerical value.

  The peak detector 48 in FIG. 6 detects a plurality of peaks PT for the average correlation value Cave calculated by the average unit 46. In part (A) of FIG. 9, the peak PT detected by the peak detector 48 from the average correlation value Cave is shown by a black circle. A known technique is arbitrarily adopted for detection of the peak PT.

  In the average correlation value Cave, a peak PT appears at each time difference L corresponding to the interval between the peaks PB of the beat index series X. On the other hand, the time point of the peak PB of the beat index series X is the time point when the intensity of the acoustic signal S changes greatly (typically, the beat point B of the music). Therefore, among the plurality of peaks PT of the average correlation value Cave, there are peaks PT that appear periodically along the time difference axis at intervals corresponding to the beat period T of the music. Therefore, the window sequence setting unit 50, the reliability calculation unit 52, and the beat period determination unit 54 in FIG. 6 are on the time difference axis for each of a plurality of intervals Δk (k = 1 to K) that are candidates for the beat period T. The probability (hereinafter referred to as “reliability”) r at which the peak PT appears in the average correlation value Cave at the interval Δk is calculated, and the interval Δk with the high reliability r is determined as the beat period T.

  By the way, the tempo of general music is in the range from 60 BPM (Beat-Per-Minute) to 200 BPM. Therefore, in this embodiment, it is assumed that the beat period T is within a range from a period corresponding to 200 BPM (0.3 s (seconds)) to a period corresponding to 60 BPM (1 s). Since the unit amount of the time difference L is 4.54 ms (milliseconds) corresponding to one beat index value OB (that is, 200 samples of the sound signal S (44.1 kHz)), the beat period corresponding to 200 BPM. (0.3 s) is 66 (= 0.3 s / 4.54 ms) when converted to time difference L, and the beat period (1 s) corresponding to 60 BPM is 220 (= 1 s / 4.54 ms) when converted to time difference L. Therefore, in this embodiment, the interval Δk is limited within the range A from 66 to 220. According to the above configuration, there is an advantage that the number of times of calculation of the reliability r (calculation amount) is reduced as compared with the case where the interval Δk is not limited.

  As shown in part (A) and part (B) of FIG. 9 (enlarged view within the range A of part (A)), the window array setting unit 50 in FIG. A window sequence for each of K peaks (PT [1] to PT [K]) of K peaks (K is a natural number) within a range A of 66 to 220 on the time difference axis from all detected peaks PT. Set GT (GT [1] to GT [K]).

  The window row GT [k] corresponding to the kth peak PT [k] is sequentially arranged on the time difference axis with an interval Δk from the peak PT [k], as shown in part (C) of FIG. This is a set of four search windows WT (WT [1] to WT [4]). The interval Δk corresponds to the time from the origin on the time axis (that is, the point where the time difference L is zero) to the peak PT [k]. Therefore, the interval Δk is different for each window row GT. Since the peak PT (PT [1] to PT [K]) used for setting the interval Δk is located in the range A, the interval Δk is limited to the range A from 66 to 220 as described above. The window width of each search window WT (WT [1] to WT [4]) is set to a predetermined value (for example, 20 beat index values OB). As shown in part (C) of FIG. 9, the search window WT [1] of the window row GT [k] is arranged so that the midpoint is located at a point separated by the interval Δk from the origin of the time difference axis. The second and subsequent search windows WT in the window row GT [k] are arranged so that the midpoint is located at a time point spaced apart from the midpoint of the immediately preceding search window WT by the interval Δk. The number of search windows WT constituting the window row GT is arbitrarily changed.

  The reliability calculation unit 52 in FIG. 6 performs the K window rows GT (GT [GT [GT [) set by the window row setting unit 50 for the K peaks PT (PT [1] to PT [K]) within the range A. 1] to GT [K]), the reliability r (r [1] to r [K]) is calculated. The reliability r [k] corresponding to the window row GT [k] is the window row GT among the plurality of peaks PT (including the peak PT outside the range A) detected by the peak detector 48 from the average correlation value Cave. [k] is an average value (or total value) of the intensities of the peaks PT located in each search window WT (WT [1] to WT [4]). When a plurality of peaks PT exist in one search window WT, the peak PT close to the midpoint of the search window WT is used to calculate the reliability r [k].

  The higher the possibility that the interval Δk of the search window WT in the window row GT [k] corresponds to the beat period T of the acoustic signal S, the greater the reliability r [k] of the window row GT [k]. The beat cycle determining unit 54 in FIG. 6 selects the window row GT [k] having the maximum reliability r from the K window rows GT (GT [1] to GT [K]), and the window row GT. The interval Δk of the search window WT at [k] is determined as the beat period T of the acoustic signal S. The beat period T specified by the beat period specifying unit 40 is stored in the storage device 14 so as to correspond to the acoustic signal S, for example. The above is the configuration and operation of the beat period specifying unit 40.

  The candidate detection unit 60 and the beat point specifying unit 70 in FIG. 1 include the beat index series X (OB [1] to OB [N]) specified by the beat index specifying unit 30 and the beat cycle T specified by the beat cycle specifying unit 40. And the beat point (beat position) B of the acoustic signal S is detected. Since the intensity of the acoustic signal S tends to change greatly at the beat point B, there is a high possibility that the peak PB corresponding to the beat point B appears in the beat index series X. However, since the peak appears in the beat index series X even when the intensity of the acoustic signal PB changes greatly except at the beat point, the peak PB cannot be simply specified as the beat point B. Candidate detection unit 60 detects M (M is a natural number) peaks PB appearing in beat index series X as shown in part (A) of FIG. The point BC (BC [1] to BC [M]) is specified and stored in the storage device 14. 1 determines the final beat point B from the M candidate points BC (BC [1] to BC [M]) specified by the candidate detection unit 60. It is not always necessary to set all the peaks PB appearing in the beat index series X as the candidate points BC of the beat point B.

  FIG. 11 is a block diagram of the beat point specifying unit 70. The window row setting unit 72 in FIG. 11 sets the window row GA (GA [1] to GA [M]) for each of the M candidate points BC (BC [1] to BC [M]) specified by the candidate detection unit 60. ) Is set. As shown in part (B) of FIG. 10, the window sequence GA [i] corresponding to the i-th (i = 1 to M) candidate point BC [i] A set of mA search windows WA (WA [1] to WA [mA]) sequentially arranged on the time axis at intervals according to. The window width w1 of each search window WA is set to a common predetermined value (for example, 20 beat index values OB) for the M window rows GA (GA [1] to GA [M]). The first search window WA [1] of the window row GA [i] is arranged so that the midpoint coincides with the candidate point BC [i], and the second search window WA [2] is the search window WA [ 1] is set so that the midpoint is located at a point separated from the midpoint (candidate point BC [i]) by the beat period T. The third and subsequent search windows WA (WA [3] to WA [mA]) are set by the following procedure.

  When the peak PB of the beat index series X does not exist in the search window WA [p-1] (p = 3 to mA) as shown in part (A) of FIG. 12, the window row setting unit 72 displays the search window WA. The search window WA [p] is set so that the time point separated from the midpoint of [p-1] by the beat period T is the midpoint. On the other hand, when the peak PB is present in the search window WA [p-1] as shown in the part (B) of FIG. 12, the window row setting unit 72 sets the time when the peak PB is separated by the beat period T from the middle. The search window WA [p] is set to be a point. When there are a plurality of peaks PB in the search window WA [p-1] as shown in part (C) of FIG. 12, the window row setting unit 72 selects the midpoint of the search window WA [p-1]. The search window WA [p] is set so that the time point separated from the peak PB close to by the beat period T is the midpoint. As shown in part (B) of FIG. 10, the arrangement of the search windows WA is repeated within a range that does not exceed a point in time before the end point (beat index value OB [N]) of the acoustic signal S by a predetermined length ΔL. The predetermined length ΔL is, for example, an added value of the beat period T and a predetermined value (for example, 10). For each of the M window rows GA (GA [1] to GA [M]), the window row setting unit 72 selects each search window WA (WA [1] to WA [mA]) of the window row GA [i]. ) Stores in the storage device 14 information (flag) indicating whether or not the peak PB of the beat index series X is included.

  11 selects the number of search windows WA including the peak PB of the beat index series X among the M window lines GA (GA [1] to GA [M]) set by the window line setting unit 72. N2 window rows GA (that is, n2 candidate points BC) having a large number are selected. As shown in FIG. 11, the selection unit 74 of this embodiment includes a first extraction unit 741 and a second extraction unit 742. The first extraction unit 741 extracts n1 window rows GA satisfying a predetermined condition from the M window rows GA (GA [1] to GA [M]), and the second extraction unit 742 includes: From the n1 window rows GA extracted by the first extraction unit 741, n2 window rows GA satisfying a predetermined condition are extracted. That is, M window rows GA (in other words, M candidate points BC (BC [1] to BC [M]) are narrowed down to n2 in two stages.

  As shown in FIG. 13, the first extraction unit 741 takes mA search windows WA (in the window row GA [i] for each of the M window rows GA (GA [1] to GA [M]). The number h1 of search windows WA including the peak PB of the beat index series X is specified among the H1 (H1 is a natural number) search windows WA selected from WA [1] to WA [mA]). The number H1 is set to a predetermined value (for example, 32). As illustrated for the window row GA [1] in the part (B) of FIG. 10, the H1 search windows WA are mA search windows WA (WA [1] to WA constituting the window row GA [i]. [mA]) is a set of search windows WA (WA [1] to WA [H1]) sequentially selected from the start point side of the acoustic signal S.

  As shown in FIG. 13, the first extraction unit 741 selects n1 window rows GA in which the number h1 exceeds a predetermined threshold TH1 among the M window rows GA (GA [1] to GA [M]). To do. The threshold TH1 is set to a value obtained by subtracting a predetermined value ΔTH1 (for example, 4) from the maximum value of the number h1 in the M window rows GA (the number h1 corresponding to the window row GA [2] in FIG. 13). In FIG. 13, it is assumed that the first extraction unit 741 selects n1 window rows GA including window rows GA [2], GA [4], GA [7], GA [9],. ing.

  As shown in FIG. 14, the second extraction unit 742 in FIG. 11 searches for the mA (all) of the window sequence GA [i] for each of the n1 window sequences GA extracted by the first extraction unit 741. Of the windows WA (WA [1] to WA [mA]), the number h2 of search windows WA that contain the peak PB of the beat index series X is specified. Further, the second extraction unit 742 selects n2 window rows GA in which the number h2 exceeds the threshold TH2 among the n1 window rows GA. The threshold TH2 is obtained by subtracting a predetermined value ΔTH2 (for example, 2) from the maximum value of the number h2 in the n1 window rows GA extracted by the first extraction unit 741 (the number h2 corresponding to the window row GA [2] in FIG. 14). It is set to the numerical value. In FIG. 14, among the n1 window rows GA (GA [2], GA [4], GA [7], GA [9],...) Extracted by the first extraction unit 741, the window row GA [2 ], GA [7],... It is assumed that the second extraction unit 742 selects n2 window rows GA including.

  The window row setting unit 76 in FIG. 11 sets a window row GB for each of n2 candidate points BC corresponding to each window row GA selected by the selection unit 74. The window sequence GB [i] corresponding to the candidate point BC [i] is represented by mB search windows WB (WB [1] to WB [mB] arranged at intervals according to the beat period T from the candidate point BC [i]. ]). Part (C) of FIG. 10 typically illustrates a window row GB [2] of candidate points BC [2]. The position of the midpoint of each search window WB (WB [1] to WB [mB]) in the window row GB [i] depends on the beat period T in the same manner as the arrangement of the search windows WA by the window row setting unit 72. Determined. As shown in part (C) of FIG. 10, the window width w2 of each search window WB in the n2 window rows GB is a predetermined value (for example, lower than the window width w1 of the search window WA set by the window row setting unit 72). 10 beat index values OB). Since the window width w1 of the search window WA and the window width w2 of the search window WB are different as described above, the position of each search window WA in the window row GA [i] and each search window WB in the window row GB [i]. Does not necessarily match the position of.

  The reliability calculation unit 78 in FIG. 11 calculates the reliability R for each of the n2 window rows GB set by the window row setting unit 76. The reliability R [i] corresponding to the window row GB [i] (candidate point BC [i]) is a numerical value serving as an index of the possibility that the candidate point BC [i] corresponds to the beat point B of the acoustic signal S. is there. For example, the reliability R [i] is the average value of the intensity (beat index value OB) of the peak PB located in each search window WB (WB [1] to WB [mB]) in the window row GB [i] or It is the total value. When a plurality of peaks PT exist in one search window WB, the peak PB close to the midpoint of the search window WB is used for calculating the reliability R.

  The beat point determination unit 80 selects any one of the n2 window rows GB (candidate points BC) selected by the selection unit 74 according to the reliability R, and the candidate corresponding to the window row GB [i]. The point BC [i] is determined as the beat point B. The peak PB that appears periodically in the beat index series X with high intensity is likely to be the beat point B. Therefore, the beat point determination unit 80 determines the candidate point BC [i] corresponding to the window row GB [i] having the highest reliability R [i] among the n2 window rows GB as the beat point B. Furthermore, the beat point determination unit 80 determines mB search windows WB (WB [1] to WB [1] to BB [i] of the candidate points BC [i] determined as the beat points B from the reliability R [i]. The time point of the peak PB included in each of WB [mB]) is determined as a series of beat points B.

  By the way, among the mB search windows WB (WB [1] to WB [mB]) of the window row GB [i] having the maximum reliability R [i], there is also a search window WB that does not include the peak PB. The beat point supplementing unit 82 in FIG. 11 calculates the interval g between the beat points B that are in succession in the series of the plurality of beat points B determined by the beat point determining unit 80, and newly adds the interval g exceeding the beat period T. A new beat point Bnew is set (supplemented). For example, as shown in part (A) of FIG. 15, when the interval g between adjacent beat points Ba and Bb is within a range of 1.5 to 2.5 times the beat period T (1.5T ≦ g <2.5T), the beat point supplementing unit 82 adds a new beat point Bnew at the midpoint between the beat points Ba and Bb. As shown in part (B) of FIG. 15, when the interval g between the beat point Ba and the beat point Bb is in the range of 2.5 to 3.5 times the beat period T (2.5T ≦ g <3.5T). ) The beat point supplementing unit 82 adds a beat point Bnew to each point that divides the interval between the beat point Ba and the beat point Bb into three equal parts.

  Further, when the peak PB of the beat index series X exists within a predetermined range with the new beat point Bnew as the midpoint, the beat point supplementing unit 82 corrects the beat point Bnew to a time corresponding to the peak PB. To do. The time point of the beat point B determined by the beat point determining unit 80 and the time point of the beat point Bnew set by the beat point supplementing unit 82 are stored in the storage device 14 so as to correspond to the acoustic signal S, for example.

  In the above embodiment, first, the selection unit 74 selects a candidate point BC corresponding to n2 window rows GA having a large number of search windows WA including the peak PB of the beat index series X, and the second Furthermore, the candidate point of the window sequence GB selected by the beat point determination unit 80 according to the intensity (beat index value OB) of the peak PB in each search window WB among the window sequences GB corresponding to the n2 candidate points BC. Confirm BC as beat point B. Since the number (n1, n2) of the search windows WA including the peak PB of the beat index series X is counted by a simple process, for example, according to this embodiment, each candidate point BC (BC [1] to BC [N ]) For each of the N window rows GA (GA [1] to GA [N]) corresponding to the intensity of the peak PB in each search window WA and calculating the beat point B Compared to the configuration to be specified, there is an advantage that the calculation amount (particularly the frequency of calculation of the reliability R) by the beat point specifying unit 70 is reduced. Further, since the window width w2 of the search window WB in the window row GB is narrower than the window width w1 of the search window WA in the window row GA, compared with the case where the search window WA and the search window WB have a common window width. It is possible to detect the beat point B accurately and efficiently.

  First, the first extraction unit 741 corresponds to the number h1 of search windows WA including the peak PB among the H1 search windows WA selected from the mA search windows WA constituting the window row GA. N1 window rows GA are selected, and secondly, the second extraction unit 742 includes the number h2 of search windows WA including the peak PB among all the search windows WA in each of the n1 window rows GA. N2 window rows GA are selected according to the above. Therefore, the window sequence corresponding to the actual beat point B is compared with the case where only the process of selecting n2 window sequences GA from the number h1 of peaks PB in the H1 search windows WA of the window sequence GA is performed. There is an advantage that GA can be accurately extracted. In addition, since the number of objects to be processed by the second extraction unit 742 is reduced to n1 in advance, all of the M window rows GA (GA [1] to GA [M]) are included in all the search windows WA. Compared with the case where n2 window rows GA are selected from the number h2 of peaks PB, there is also an advantage that the calculation amount by the selection unit 74 is reduced.

  When the window row setting unit 72 arranges each search window WA, the next search window WA is set with the point when the beat period T has elapsed from the point of the peak PB included in the search window WA as a midpoint. For example, compared to a configuration in which the starting point of the beat period T is fixed at the midpoint of each search window WA, each beat point B can be accurately detected even when there is fluctuation in the beat period T of the acoustic signal S. There are advantages. Although the search window WA has been described above, the same effect can be realized for the setting of the search window WB by the window row setting unit 76.

  When the interval g between the beat points B determined by the beat point determination unit 80 exceeds the beat period T, a new beat point Bnew is supplemented within the interval g, so that the intensity change in the acoustic signal S is relatively small. There is an advantage that a beat point B (that is, a beat point B that does not appear as a prominent peak PB in the beat index series X) can also be specified appropriately.

<Modification>
Various modifications are added to the above embodiments. An example of a specific modification is as follows. In addition, you may combine each aspect illustrated below arbitrarily.

(1) Modification 1
The method for specifying the beat period T is not limited to the above examples. For example, in the above embodiment, the average correlation value Cave obtained by averaging the autocorrelation value C for each unit interval U with respect to NU unit intervals U is used for specifying the beat period T. However, the autocorrelation of a specific unit interval U is used. A configuration in which the value C is used instead of the average correlation value Cave (a configuration in which the average unit 46 is omitted) is also employed. However, in the autocorrelation value C of one unit section U, the influence of factors other than the beat point B (for example, noise) becomes obvious, so that there is a problem that the specific accuracy of the beat period T is limited. In the configuration of FIG. 6, an average correlation value Cave over NU unit intervals U is used, so that the influence of factors other than the beat point B in the acoustic signal S can be reduced and the beat period T can be specified with high accuracy. There is.

  In addition to the above examples, a known technique is arbitrarily employed for specifying the beat period T. Moreover, the process which calculates the beat period T by the calculation with respect to the acoustic signal S is not essential in this invention. For example, in the configuration in which the beat period T of the acoustic signal S is stored in the storage device 14, the process of reading the beat period T from the storage device 14 corresponds to “specification of the beat period” in the present invention. However, in the configuration of FIG. 6, the beat index series X specified by the beat index specifying unit 30 is used both for specifying the beat period T by the beat period specifying unit 40 and for specifying the beat point B by the beat point specifying unit 70. Therefore, as compared with a configuration in which the beat period specifying unit 40 does not use the beat index series X to specify the beat period T, there is an advantage that the calculation amount of the sound processing apparatus 100 as a whole is reduced.

(2) Modification 2
The definition and calculation method of the beat index series X (X0) are changed as appropriate. For example, in the above description, the beat index series X is calculated using the change in intensity over the entire band of the acoustic signal S of each frame F as the beat index value OA, but the acoustic signal S has a specific frequency band (for example, a high frequency range). A configuration for calculating the beat index series X from the intensity of the component to which it belongs, and a structure for calculating the beat index series X using the relative ratio of the intensity of the acoustic signal S in each successive frame F as the beat index value OA are also employed.

  Further, the feature quantity that is the basis of the beat index series X is not limited to the intensity of the acoustic signal S. For example, the beat index series X is calculated using the phase deviation of the complex spectrum of each frame F as the beat index value OA. More specifically, the difference between the phase of the short-time complex spectrum of the acoustic signal S predicted for each frame F from the past frame F and the actual phase of the frame F (or simply the phases of the preceding and subsequent frames F). The beat index series X is generated by calculating a value obtained by summing or averaging the absolute values of the difference values over the entire band as the beat index value OA of each frame F. Even when the phase deviation is used, the position of the beat point B (striking sound) can be detected with high accuracy.

  Further, a configuration in which the beat index series X0 is specified using a distance (for example, Euclidean distance) between vectors specified from the frequency spectrum Q of each successive frame F as a beat index value OA, and each successive frame in the acoustic signal S. A configuration (a configuration in which the calculation of the frequency spectrum Q is omitted) that calculates the time series of the difference in F intensity (volume) as the beat index sequence X0 is also employed. As can be understood from the above examples, any numerical value serving as an index of the degree of change in the characteristic amount of the acoustic signal S is used as the beat index value in the present invention.

  A configuration in which the peak emphasizing unit 34 is omitted is also suitable. That is, the series of beat index values OA (OA [1] to OA [N]) calculated by the index calculation unit 32 is used as the beat index series X by the beat period specifying unit 40 and the candidate detection unit 60. In the configuration including the peak emphasizing unit 34, the method for emphasizing the peak of the beat index series X0 is not limited to the above example. For example, the beat index series X is specified by performing high-pass filter processing on the beat index series X0.

(3) Modification 3
In the above embodiment, the candidate point BC (window row GA) is narrowed down by the first extraction unit 741, the candidate point BC (window row GA) is narrowed down by the second extraction unit 742, and the candidate point BC corresponding to the reliability R is selected. Although the configuration for performing the narrowing (determination of beat point B) is illustrated, a configuration that employs only two arbitrary types of narrowing out of the above three types of narrowing down is also suitable. For example, a configuration in which one of the first extraction unit 741 and the second extraction unit 742 is omitted or a configuration in which the reliability calculation unit 78 (and further the window row setting unit 76) is omitted is also employed. In the configuration in which the reliability calculation unit 78 is omitted, for example, the window row GA [i [i] in which the number h2 of search windows WA including the peak PB among the n1 window rows GA selected by the first extraction unit 741 is maximum. ] Corresponding to the point BC [i] is determined as the beat point B.

(4) Modification 4
In the above embodiment, the second extraction unit 742 specifies the number h2 for all (mA) search windows WA (WA [1] to WA [mA]) in the window row GA [i]. Only a part of the search windows WA in the window row GA [i] may be used for counting the number h2. That is, in the configuration including the first extraction unit 741 and the second extraction unit 742, the number H2 of search windows WA used for counting the number h2 by the second extraction unit 742 is the number by the first extraction unit 741. A configuration in which the number of search windows WA used for counting h1 exceeds the number H1 is suitable, and it does not matter whether the number of search windows WA is all or part of the window row GA.

  The H1 search windows WA used by the first extraction unit 741 for counting the number h1 are arbitrarily selected from the window row GA. For example, a configuration for selecting H1 search windows WA corresponding to the middle part or the last part of the acoustic signal S, or a configuration for selecting H1 search windows WA in a distributed manner from the start point to the end point of the acoustic signal S. Is adopted. However, since beat point B (beat interval T) tends to appear clearly and stably in the first part of the music, there are many H1 search windows WA used for counting the number h1. According to the configuration in which the acoustic signal S (musical piece) is selected in order from the start point side, there is an advantage that the beat point B can be detected accurately and stably as compared with the case where the search window WA of another part is used.

(5) Modification 5
Although the case where the beat period T (tempo) is constant over the entire acoustic signal S has been exemplified in the above form, even if the beat period T changes in each part of the acoustic signal S, the above form is adopted. The specified beat period T and beat point B are applied. FIG. 16 is a conceptual diagram illustrating the autocorrelation value C for each unit section U when the beat period T changes in the music piece in the same manner as the part (A) of FIG. Since the autocorrelation value C has a peak (PT) at intervals corresponding to the beat period T, the peak of the autocorrelation value C is obtained in each section V (V1, V2, V3) having a different beat period T in the music. The interval of PT is different. FIG. 17 is a block diagram of the beat period specifying unit 40 according to the modification. 17 calculates the interval of the peak PT of the autocorrelation value C in each unit interval U, and divides the acoustic signal S into a plurality of intervals V with the time when the interval changes as a boundary. Each element from the averaging unit 46 to the beat period determining unit 54 in FIG. 17 specifies the beat period T for each section V by executing the same processing as in the above form (FIG. 1) for each of the plurality of sections V. To do. On the other hand, the beat point specifying unit 70 uses the beat index series X in the section V and the beat period T in the section V for each of the plurality of sections V of the acoustic signal S, and uses the same method as the above embodiment. The beat point B series is identified with. In addition, the method of detecting the change of the beat period T is not limited to the above illustration.

(6) Modification 6
In the above embodiment, the window row GB [i] is used to calculate the reliability R [i]. However, the reliability calculation unit 78 uses the window row GA [i] set by the window row setting unit 72 to perform the trust. A configuration for calculating the degree R [i] is also employed. For example, the reliability calculation unit 78 includes, for each of the n2 window rows GA selected by the selection unit 74, within each search window WA (WA [1] to WA [mA]) of the window row GA [i]. The average value or total value of the intensities (beat index value OB) of the peak PB located is calculated as the reliability R [i]. As described above, the window row setting unit 76 is omitted in the configuration in which the window row GA [i] is used for calculating the reliability R [i]. However, the window row GB in which the search windows WB having a narrow window width w2 compared to the window width w1 of the search window WA as shown in the parts (B) and (C) of FIG. According to the configuration used for the calculation, the possibility that the peak PB appears in the search window WB due to factors other than the beat point B (for example, noise) is reduced. Compared with the configuration in which the window row setting unit 76 is omitted. Thus, there is an advantage that the beat point B can be accurately detected.

(7) Modification 7
In the above embodiment, the beat point B over the entire acoustic signal S is detected, but a configuration in which only the first beat point B (candidate point BC) of the music is detected is also employed. For example, the beat point determining unit 80 determines only the candidate point BC [i] corresponding to the window row GB [i] having the maximum reliability R [i] as the first beat point B of the music. Moreover, the structure which abbreviate | omitted the beat point supplement part 82 is also employ | adopted.

(8) Modification 8
In the above embodiment, the configuration in which the acoustic signal S is stored in the storage device 14 is illustrated. However, in the configuration in which the frequency spectrum Q of each frame F of the acoustic signal S is calculated in advance and stored in the storage device 14, The acoustic signal S need not be held in the storage device 14.

(9) Modification 9
In the above embodiment, the sound processing device 100 including the beat index specifying unit 30, the beat cycle specifying unit 40, and the beat point specifying unit 70 is exemplified. However, the beat index specifying unit 30, the beat cycle specifying unit 40, and the beat point are exemplified. Each of the specifying units 70 is established as an independent device. That is, the sound processing apparatus including the beat index specifying unit 30 specifies the beat index series X (beat index values OB [1] to OB [N]) used for specifying the beat period T and detecting the beat point B. The sound processing apparatus including the beat period specifying unit 40 is established as an apparatus for specifying the beat period T (tempo) from the beat index series X (beat index values OB [1] to OB [N]). The sound processing device including the beat point specifying unit 70 is established as a device that specifies the beat point B from the beat index series X (beat index values OB [1] to OB [N]) and the beat period T.

1 is a block diagram of a sound processing apparatus according to an embodiment of the present invention. It is a conceptual diagram for demonstrating each flame | frame of an acoustic signal. It is a block diagram of the beat index specific part in FIG. It is a conceptual diagram for demonstrating a beat parameter | index series and a unit area. It is a conceptual diagram for demonstrating the smooth value used in a peak emphasis part. It is a block diagram of the beat period specific | specification part in FIG. It is a graph which shows the autocorrelation value of a beat index series. It is a conceptual diagram for demonstrating calculation of an average correlation value. It is a conceptual diagram for demonstrating the setting of the window row | line | column in a beat period specific | specification part. It is a conceptual diagram for demonstrating operation | movement of the candidate detection part and beat point specific | specification part in FIG. It is a block diagram of the beat point specific | specification part in FIG. It is a conceptual diagram for demonstrating operation | movement of a window row setting part. It is a conceptual diagram for demonstrating operation | movement of a 1st extraction part. It is a conceptual diagram for demonstrating operation | movement of a 2nd extraction part. It is a conceptual diagram for demonstrating operation | movement of a beat point supplement part. It is a conceptual diagram which shows the autocorrelation value of each unit area in case a beat cycle changes. It is a block diagram of the beat period specific part concerning a modification.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Acoustic processing apparatus, 12 ... Control apparatus, 14 ... Memory | storage device, 20 ... Frequency analysis part, 30 ... Beat index specific | specification part, 32 ... Index calculation part, 34 ... Peak emphasis part, 40 ... ... Beat period specifying part, 42 ... Section setting part, 44 ... Correlation calculating part, 46 ... Average part, 48 ... Peak detecting part, 50 ... Window row setting part, 52 ... Reliability calculating part, 54 ...... Beat cycle determination unit, 56 …… Section detection unit, 60 …… Candidate detection unit, 70 …… Beat point identification unit, 72 …… Window sequence setting unit, 74 …… Selection unit, 741 …… First extraction unit 742 ... second extraction unit 76 ... window row setting unit 78 ... reliability calculation unit 80 ... beat point determination unit 82 ... beat point supplementation unit

Claims (12)

A beat index specifying means for specifying a time series of beat index values indicating changes in the characteristic amount of the acoustic signal;
Beat period specifying means for specifying a beat period of the acoustic signal;
Candidate detection means for detecting a plurality of peaks in the time series of the beat index value;
Each of the plurality of peaks as beat point candidate points, a first window row setting means for setting a first window row in which a plurality of search windows are arranged from the candidate points at intervals according to the beat period;
Selecting means for selecting candidate points corresponding to two or more window rows having a large number of search windows including a peak of the beat index value among the plurality of first window rows;
Of the two or more candidate points selected by the selection means, a candidate point selected according to the peak beat index value in each search window arranged at an interval corresponding to the beat cycle from the candidate point is determined as a beat point. A sound processing apparatus comprising: beat point determination means. For each of the two or more candidate points selected by the selection means, a plurality of search windows having a window width narrower than the window width of each search window in the first window row are extracted from the candidate points at intervals corresponding to the beat period. Comprising second window row setting means for setting the arranged second window rows;
The sound processing device according to claim 1, wherein the beat point determination unit determines a candidate point selected according to a peak beat index value in each search window from the plurality of second window rows as a beat point. The selection means includes
Among the first number of search windows in the first window row, a plurality of first window rows having two or more search windows including a peak of the beat index value are set by the first window row setting means. First extraction means for selecting from the first window row;
The first extracting means selects two or more first window rows having a large number of search windows including a peak of a beat index value among a second number of search windows exceeding the first number in the first window row. The sound processing apparatus according to claim 1, further comprising: a second extraction unit that selects the detected two or more first window rows. A beat index specifying means for specifying a time series of beat index values indicating changes in the characteristic amount of the acoustic signal;
Beat period specifying means for specifying a beat period of the acoustic signal;
Candidate detection means for detecting a plurality of peaks in the time series of the beat index value;
Each of the plurality of peaks as beat point candidate points, a first window row setting means for setting a first window row in which a plurality of search windows are arranged from the candidate points at intervals according to the beat period;
Among the first number of search windows in the first window row, a plurality of first window rows having two or more search windows including a peak of the beat index value are set by the first window row setting means. First extraction means for selecting from the first window row;
The first extraction means detects the first window row in which the number of search windows including the peak of the beat index value is the maximum among the second number of search windows exceeding the first number in the first window row. And a second extraction unit that selects from the two or more first window rows and determines a candidate point corresponding to the first window row as a beat point. 5. The acoustic processing device according to claim 3, wherein the first number of search windows is a plurality of search windows sequentially selected from a start point side of the acoustic signal among the plurality of search windows of the first window row. . The sound processing device according to any one of claims 1 to 5, wherein the first window row setting means sets the next search window with the time when the beat period has elapsed from a peak included in one search window as the center of the window width. . The sound according to any one of claims 1 to 3 , wherein the beat point determining means detects, as beat points, peaks in each search window arranged at intervals according to the beat period from candidate points determined as beat points. Processing equipment. The sound processing device according to claim 7, further comprising beat point supplementing means for setting a new beat point between the beat points when the interval between the beat points determined by the beat point determining means exceeds the beat cycle. . The beat period specifying means includes
Correlation calculating means for calculating an autocorrelation value for each unit section dividing the time series of the beat index values;
9. An average means for averaging the autocorrelation values calculated by the correlation calculation means for a plurality of unit sections, wherein the beat period is specified from the autocorrelation values after the averaging by the averaging means. Sound processing device. The beat index specifying means includes
Index calculation means for calculating a time series of beat index values indicating changes in the characteristic amount of the acoustic signal;
The sound processing apparatus according to claim 1, further comprising: a peak emphasizing unit that emphasizes an increase / decrease in a beat index value in a time series of beat index values calculated by the index calculating unit. A beat index specifying process for specifying a time series of beat index values indicating changes in the characteristic amount of the acoustic signal;
A beat period specifying process for specifying a beat period of the acoustic signal;
Candidate detection processing for detecting a plurality of peaks in the time series of the beat index value;
Each of the plurality of peaks as beat point candidate points, a first window row setting process for setting a first window row in which a plurality of search windows are arranged from the candidate points at intervals according to the beat period;
A selection process for selecting candidate points corresponding to two or more window rows having a large number of search windows including the peak of the beat index value among the plurality of first window rows;
Of the two or more candidate points selected in the selection process, the candidate point selected according to the beat index value of the peak in each search window arranged at intervals according to the beat period from the candidate point is determined as the beat point. A program that causes a computer to execute beat point determination processing. A beat index specifying process for specifying a time series of beat index values indicating changes in the characteristic amount of the acoustic signal;
A beat period specifying process for specifying a beat period of the acoustic signal;
Candidate detection processing for detecting a plurality of peaks in the time series of the beat index value;
Each of the plurality of peaks as beat point candidate points, a first window row setting process for setting a first window row in which a plurality of search windows are arranged from the candidate points at intervals according to the beat period;
Among the first number of search windows in the first window row, a plurality of two or more first window rows having a large number of search windows including the peak of the beat index value are set in the first window row setting process. A first extraction process for selecting from the first window row;
The first extraction process detects the first window row having the maximum number of search windows including the peak of the beat index value among the second number of search windows exceeding the first number in the first window row. A second extraction process for selecting from the two or more first window rows and determining candidate points corresponding to the first window rows as beat points;
A program that causes a computer to execute.

Images