JP2007272118A - Method, device, and program for extracting musical piece tempo - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To accurately extract a tempo without being influenced by a peak detection error which is likely to occur on a sampled voice data. <P>SOLUTION: Two or more peaks are detected from the sampled voice data of an arbitrary musical piece (Step S14). Based on a 1st peak train before movement located in a line according to the peak generation time of these two or more detected peaks, and a 2nd peak train after the peak generation time of the 1st peak train are moved by a prescribed time (t), a distance function between peaks used as a function of the time (t) is calculated (Step S16) which is the total sum of time between respectively nearest peaks between the 1st peak train and the 2nd peak train. Since this distance function between peaks changes periodically according to the periodicity of intervals of the 1st peak train, the tempo of the arbitrary musical piece is calculated based on the cycle of the distance function between peaks (Step S18). <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a music tempo extraction method, apparatus, and program, and more particularly to a technique for extracting the tempo of music from sampling audio data of an arbitrary music.

  2. Description of the Related Art Conventionally, techniques for analyzing digital audio data storing music and automatically calculating various feature values of the music such as tempo / rhythm / accent have been widely developed.

  Among them, among the technologies for automatically extracting “tempo”, which is one of the feature quantities of music, not so-called music score information format music data (non-waveform data) such as MIDI but sampled audio data such as WAV / MP3 is used. As what is handled, there are those described in Patent Documents 1 to 3.

  The tempo extraction device described in Patent Document 1 calculates a measure time length from audio data, and calculates the tempo based on the calculated measure time length and the number of beats per measure. However, the condition that the user specifies the number of beats per measure is necessary.

  The tempo data generation method described in Patent Document 2 selects two optimal peak values, and calculates the tempo from the interval between the selected peak values. An algorithm for selecting the optimal two peak values There is a problem with the accuracy of the tempo calculation because there is an unclear point in the legitimacy, and since attention is paid only to adjacent peaks.

The beat number detection device described in Patent Document 3 calculates the tempo using the interval between the peak positions of the low frequency components, but only for songs with a tempo (BPM (Beat Per Minute)) of 100 to 150. Yes.
JP-A-5-27751 Japanese Patent Laid-Open No. 7-64544 JP-A-8-201542

  Each of the techniques described in Patent Documents 1 to 3 detects a peak from waveform data in a certain frequency band, and obtains a tempo from the distance (time interval) between adjacent peaks. This is a general technique of tempo extraction technology.

  Since these are calculated on the assumption that the distance between the peaks is an integer ratio, there is a drawback that when the deviation occurs in the peak position, the deviation gradually increases when the calculation is performed with the integer ratio.

  In general, since many sounds are mixed in sampled sound data, an error is likely to occur at the time of peak detection. In the conventional method, there is a high possibility that the tempo cannot be accurately detected due to the peak detection error that is likely to occur in the sampled sound data.

  Many conventional methods limit the conditions (musical instrument, frequency, tempo) of the target sound source to some extent.

  The present invention has been made in view of such circumstances, and can accurately extract the tempo without being affected by the peak detection error that is likely to occur in the sampled sound data, and has high robustness without specifying the target sound source. It is an object of the present invention to provide a music tempo extraction method, apparatus, and program that can be realized.

  In order to achieve the above object, a music tempo extraction method according to claim 1 includes: an audio data input step for inputting sampling audio data of an arbitrary music; and detecting a plurality of peaks of the input sampling audio data, and detecting the peaks A peak detection step for detecting a peak occurrence time of the plurality of peaks, a first peak row before movement in which the detected plurality of peaks are arranged in the peak occurrence time, and a peak occurrence time of the first peak row. Based on the second peak train after movement that has been moved for a predetermined time (t), it is the sum of the times between the nearest peaks between the first peak train and the second peak train, A peak-to-peak distance function calculating step for calculating a peak-to-peak distance function that is a function of the time (t), and a template of the arbitrary music piece based on the calculated period of the peak-to-peak distance function. It is characterized in that it comprises a tempo calculation step of calculating a.

  That is, a plurality of peaks arranged in order of time are detected from sampling audio data of an arbitrary music piece. Then, a peak-to-peak distance function newly introduced in the present invention is calculated based on a peak row in which a plurality of peaks are arranged at peak occurrence times. This inter-peak distance function moves a peak sequence before movement (first peak sequence) in which a plurality of peaks are arranged in the peak occurrence time and a peak occurrence time of the first peak sequence by a predetermined time (t). Based on the peak sequence after movement (second peak sequence), it is the sum of the time between the nearest peaks between the first peak sequence and the second peak sequence. The inter-peak distance function is a function of the time (t), and changes periodically according to the periodicity of the interval between the first peak rows. Therefore, the tempo of the arbitrary music can be calculated based on the period of the peak-to-peak distance function. According to this method, even if there is a slight error in peak detection, there is no effect on the period of the inter-peak distance function calculated based on the information of a plurality of peaks, and the tempo can be accurately extracted.

  According to a second aspect of the present invention, in the music tempo extraction method according to the first aspect, the peak detection step detects a peak value of each peak together with a peak occurrence time of the plurality of peaks, and sets the detected peak value. On the basis of this, it is characterized in that those having a small peak value are removed from the first peak row. That is, before applying to the peak-to-peak distance function, filtering is performed by the peak intensity. This removes weak peaks and reduces noise to improve the accuracy of tempo calculation.

  3. The music tempo extraction method according to claim 1 or 2, wherein the peak detection step automatically selects a band in which the tempo can be accurately obtained from the sampled sound data before detecting the peak. It is characterized by doing. As a result, data of the sound band of the musical instrument that regularly rhythms is extracted, and the accuracy of subsequent tempo extraction can be improved.

  4. The music tempo extraction method according to claim 1 or 2, wherein the peak detecting step is a step of filtering the sampled audio data into a plurality of divided bands before detecting the peak. Calculating the autocorrelation function of the filtered data of each band, and selecting the sampling audio data of the band from which the most regular autocorrelation function is obtained among the calculated autocorrelation functions. It is a feature.

  5. The music tempo extraction method according to claim 1, wherein the tempo calculation step includes the time (t) when the calculated peak-to-peak distance function first becomes a minimum value. ) Or the period of the peak-to-peak distance function based on the time between adjacent local minimum values.

  6. The music tempo extraction method according to claim 1, wherein the tempo calculation step determines whether the calculated tempo is a double tempo or a half tempo. Including a step and correcting the tempo based on the determination result.

  The music tempo extraction device according to claim 7 is an audio data input means for inputting sampling audio data of an arbitrary music, and detects a plurality of peaks of the sampling audio data, and calculates peak occurrence times of the detected peaks. Peak detecting means for detecting, the first peak train before movement in which the detected plurality of peaks are arranged in the peak occurrence time, and the peak occurrence time of the first peak train are moved by a predetermined time (t) Based on the second peak train after movement, it is a function of the time (t), which is the sum of the times between the nearest peaks between the first peak train and the second peak train. A peak-to-peak distance function calculating unit that calculates a peak-to-peak distance function; and a tempo calculating unit that calculates a tempo of the arbitrary music based on a cycle of the calculated peak-to-peak distance function. It is a symptom.

  A music tempo extraction program according to claim 8 causes a computer to realize the music tempo extraction method according to any one of claims 1 to 6.

  According to the present invention, it is possible to accurately extract a tempo without being affected by a peak detection error that is likely to occur in sampled sound data.

  Preferred embodiments of a music tempo extraction method, apparatus, and program according to the present invention will be described below with reference to the accompanying drawings.

  FIG. 1 is a block diagram showing a hardware configuration example of a personal computer having a function as a music tempo extraction device according to the present invention.

  As shown in FIG. 1, a personal computer 1 includes a central processing unit (CPU) 10 that mainly controls the operation of each component, a main memory 12 that stores a control program for the device, and serves as a work area when the program is executed. Operating system (OS) of the personal computer 1, device drivers of peripheral devices connected to the personal computer 1, various application software including the music tempo extraction program according to the present invention, sampling audio data such as WAV / MP3, etc. A device 14, a CD-ROM device 16, a display memory 18 for temporarily storing display data, a CRT monitor or a liquid crystal monitor for displaying images, characters, etc. by image data, character data, etc. from the display memory 18. Monitor device 20, sound card 22, speaker 24, keyboard 26, a mouse 28 as a position input device, a mouse controller 30 for detecting the state of the mouse 28 and outputting signals such as the position of the mouse pointer on the monitor device 20 and the state of the mouse 28 to the CPU 10, and a USB (Universal Serial) Bus), and a bus 34 for connecting the above components.

  The personal computer 1 having the above-described configuration is well-known except for the music tempo extraction program according to the present invention, which is stored in the hard disk device 14, and a detailed description of each component will be omitted. This music tempo extraction program can be installed in the personal computer 1 by setting the CD-ROM on which the program is recorded in the CD-ROM device 16 of the personal computer 1 or through a network not shown.

  FIG. 2 is a functional block diagram showing the function of the personal computer 1 as a music tempo extraction device.

  As shown in the figure, the music tempo extraction device includes a music data input device 50, a storage device 52, an operation means 54, and a tempo extraction means 56.

  The music data input device 50 is a means for inputting music data (sampling audio data) of a music to be subjected to tempo extraction. For example, the CD-ROM device 16 of FIG.

  The storage device 52 is for temporarily storing the input sampling sound data, and the main memory 12 and the hard disk device 14 of FIG. 1 are conceivable.

  The operation means 54 receives an operation from the user, selects a music piece from which a tempo is extracted, and gives an instruction for tempo extraction. The keyboard 26 and the mouse 28 in FIG.

  The tempo extraction means 56 is means for executing tempo extraction according to the present invention, and a music tempo extraction program stored in the hard disk device 14 of FIG. 1 and a CPU 10 for executing the program can be considered.

  Next, the tempo extraction process in the tempo extraction means 56 will be described in detail.

  FIG. 3 is a flowchart mainly showing the flow of processing of the tempo extracting means 56.

  In the figure, a tempo extraction means 56 inputs music data to be subjected to tempo extraction (step S10). The music data to be input is not non-waveform data having musical score information such as MIDI but sampling audio data such as WAV / MP3 as shown in FIG. In this embodiment, sampling audio data for 10 or more seconds is captured.

  Subsequently, the tempo extraction means 56 smoothes the input sampling sound data (step S12). FIG. 5 is a graph showing data on the plus side obtained by smoothing the sampled sound data. The smoothing method may be a known method such as a moving average method (a method for calculating an average value of points included in a certain section).

  Note that the waveform data of the input sound source may be filtered as necessary in order to remove unnecessary band data at a timing before smoothing and improve the accuracy thereafter. The filtering technique may be a known technique using FFT (Fourier transform), FIR, IIR filter, or the like.

  Next, the tempo extraction means 56 detects the peak of the smoothed waveform data as a peak (step S14). FIG. 6 shows peak occurrence positions (time) and peak intensities (peak values) of a plurality of peaks detected during a few tens of seconds.

  This detected peak is filtered by the intensity of the peak before being subjected to the “inter-peak distance function” described later. This filtering removes a weak peak (a weak peak surrounded by a circle A in FIG. 6). This filtering method may be a method of removing a peak having a peak value below a certain threshold value, or a method of picking up a peak satisfying a certain number in order from the strongest one.

  Subsequently, the tempo extraction means 56 performs a calculation process necessary for tempo extraction based on the detected peak, using a unique “inter-peak distance function” (step S16). The method used here is original and is a characteristic point of the present invention.

  Next, the “peak distance function” will be described.

  For example, assume that a peak is detected as shown in FIG. Here, in order to simplify the description, it is assumed that the number of peaks is three and the peaks are arranged at approximately equal intervals.

  When this peak row is moved to the right on FIG. 7 by a predetermined movement amount (time) t (= 0.1 seconds) while maintaining the positional relationship (time interval) of these three peaks (peak row). FIG. 7B shows the relationship between the peak train before movement (first peak train) and the peak train after movement (second peak train).

  Then, the “total sum of distances” between the nearest peaks is calculated between the first peak row and the second peak row. In this case, “total distance” = 0.1 × 3 = 0.3 (sec). Here, “distance” means time.

  If the movement amount t is increased little by little, a certain peak after movement may approach another peak before movement. FIG. 7C shows a case where the second peak row moves to the right by 1.0 second with respect to the first peak row.

  As described above, when the nearest peak becomes the adjacent peak as a result of moving the second peak row, the “distance” with the adjacent peak is adopted, and the “total distance” is calculated. In this case, “sum of distances” = (3.1-2.5) + (4.6-4.1) + (5.6-4.6) = 0.6 + 0.5 + 1.0 = 2. 1 (sec).

  Thus, the “total distance” with respect to the movement amount t when the second peak row is moved is referred to as a peak-to-peak distance function PD (t).

  If there is a period at the peak position of the peak row, the peak-to-peak distance function PD (t) also has a period.

  FIG. 8 shows the peak-to-peak distance function PD (t) calculated as described above, and the distance between the peaks is constant at 0.5 (sec) (the horizontal axis in FIG. 8). Is m sec).

  Finally, the tempo extraction means 56 calculates the tempo of the music from the peak-to-peak distance function calculated as described above (step S18).

  Assuming that the continuity of the peaks is regular, it can be expected that the point-to-peak distance function becomes minimal when the peaks are shifted and is the same as the original sequence of peaks. In other words, since the movement amount t from the start until the peak-to-peak distance function is minimized should be a beat unit, the tempo can be calculated from the beat unit section and the real time.

It can be seen that the peak-to-peak distance function shown in FIG. 8 has a minimum value with a period of 0.5 (sec). If the tempo is calculated using this period 0.5 (sec), it can be seen that the tempo (BPM) can be calculated by the following equation, and a correct result is obtained.
[Equation 1]
Tempo = 60 ÷ 0.5 = 120
In addition, as a period of the distance function between peaks, you may obtain | require from the movement amount t from which a distance function between peaks becomes a minimum value first, and may obtain | require from the space | interval of an adjacent minimum value.

  In the music tempo extraction method according to the present invention, the entire peak row is slid as one lump, and the regularity of the whole lump is judged. Therefore, errors (positional deviations) occurring in individual peaks do not affect the accuracy. . In addition, this method does not need to specify conditions such as the frequency band, tempo, and time signature that are limited by the prior art for the target music data, and can exhibit high robustness.

  Next, a method for automatically selecting waveform data in a band in which the tempo can be most accurately detected in the peak detection step shown in FIG. 3 will be described.

  As shown in the flowchart of FIG. 9, first, music data to be subjected to tempo extraction is input (step S20). This step S20 is the same as step S10 shown in FIG.

  Subsequently, the waveform data of the entire music is filtered by paying attention to a certain band (step S22). For example, in one octave unit, several types of band data from high to low range are filtered.

  Next, an autocorrelation function (ACF) is applied to the filtered waveform data of each band (step S24).

  By comparing the autocorrelation functions of the respective bands obtained in step S24, the band in which the most regular autocorrelation function (having a large amplitude and similar repeating components) is obtained is used as a band used for tempo extraction. Determine (step S28).

  Thereby, for example, data of a sound band of a musical instrument that regularly rhythms such as a drum is extracted, and the accuracy of subsequent tempo extraction can be improved.

  Next, a method for preventing a tempo detection error such as a double tempo and a half tempo and correcting the tempo correctly in the tempo calculation step shown in FIG. 3 will be described.

  As shown in the flowchart of FIG. 10, tempo calculation is performed in step S30. In this tempo calculation step S30, the tempo is calculated based on the result of the peak-to-peak distance function as in step S18 of FIG.

  Next, it is determined whether the calculated tempo is a double tempo or a half tempo, and the tempo is corrected based on the determination result (step S32).

  This correction method is arbitrary. For example, as in the method described in Japanese Patent Application Laid-Open No. 2005-274708, a value of speed sensation S of music is obtained, and when the value is higher than the upper limit threshold value, that is, In the case of music with a sense of speed, the tempo is corrected twice. On the other hand, when the music is lower than the lower threshold, that is, in the case of music without a sense of speed, the tempo is corrected to a half. Perform the operation.

  Depending on the correction method, there may be a case where another audio feature quantity such as the sense of speed S is required, and the feature quantity is calculated in advance at an appropriate timing of tempo extraction, so that the entire calculation time is calculated. You may devise to reduce.

  Further, in this embodiment, the sampling voice data of 10 or more seconds is input, but the length of the input sampling voice data is not limited to this, and the point is that the influence of the peak detection error does not occur. Any length that can detect a number of peaks is acceptable. Furthermore, in this embodiment, the case of extracting the tempo of music by software has been described using a personal computer as an example, but may be realized by hardware.

FIG. 1 is a block diagram showing a hardware configuration example of a personal computer provided with a function as a music tempo extraction device according to the present invention. FIG. 2 is a functional block diagram showing functions as a music tempo extraction device in the personal computer. FIG. 3 is a flowchart showing the flow of processing of the tempo extracting means shown in FIG. FIG. 4 is a graph showing an example of input sampling sound data. FIG. 5 is a graph showing an example of the smoothed sampling audio data. FIG. 6 is a graph showing peak occurrence positions (time) and peak intensities (peak values) of a plurality of detected peaks. FIG. 7 is a diagram used for explaining the peak-to-peak distance function. FIG. 8 is a graph showing an example of the peak-to-peak distance function. FIG. 9 is a flowchart used to explain a method of automatically selecting waveform data in a band in which the tempo is most easily detected. FIG. 10 is a flowchart used to explain a method of correcting the double tempo, half tempo, and the like.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Personal computer, 10 ... Central processing unit (CPU), 12 ... Main memory, 14 ... Hard disk drive, 16 ... CD-ROM, 20 ... Monitor device, 26 ... Keyboard, 28 ... Mouse, 50 ... Music data input device, 52 ... Storage device 54 ... Operating means 56 ... Tempo extracting means

Claims (8)

An audio data input step for inputting sampling audio data of an arbitrary song;
A peak detection step of detecting a plurality of peaks of the input sampling voice data and detecting a peak occurrence time of the detected plurality of peaks;
The first peak row before movement in which the detected plurality of peaks are arranged in the peak occurrence time, and the second peak after movement obtained by moving the peak occurrence time of the first peak row by a predetermined time (t) Based on the column, a peak-to-peak distance function that is a function of the time (t), which is the sum of the time between the nearest peaks between the first peak column and the second peak column, is calculated. A peak-to-peak distance function calculation step;
A tempo calculation step of calculating the tempo of the arbitrary music based on the calculated period of the peak-to-peak distance function;
The music tempo extraction method characterized by including this.   The peak detection step detects a peak value of each peak together with the peak occurrence times of the plurality of peaks, and removes a small peak value from the first peak row based on the detected peak value. The music tempo extraction method according to claim 1.   3. The music tempo extraction method according to claim 1, wherein the peak detection step automatically selects a band from which the tempo can be accurately obtained from the sampled audio data before detecting the peak.   The peak detection step includes: filtering the sampled audio data into a plurality of divided bands before detecting the peak; calculating an autocorrelation function of the data of each filtered band; The music tempo extraction method according to claim 1 or 2, wherein sampling audio data in a band in which the most regular autocorrelation function is obtained among the calculated autocorrelation functions is selected.   The tempo calculation step obtains the period of the peak-to-peak distance function based on the time (t) when the calculated peak-to-peak distance function first becomes a minimum value, or the time between adjacent minimum values. The music tempo extraction method according to any one of claims 1 to 4.   6. The tempo calculation step includes a step of determining whether the calculated tempo is a double tempo or a half tempo, and the tempo is corrected based on the determination result. The music tempo extraction method according to any one of the above. Audio data input means for inputting sampling audio data of an arbitrary song;
A peak detecting means for detecting a plurality of peaks of the sampling voice data and detecting a peak occurrence time of the detected plurality of peaks;
The first peak row before movement in which the detected plurality of peaks are arranged in the peak occurrence time, and the second peak after movement obtained by moving the peak occurrence time of the first peak row by a predetermined time (t) Based on the column, a peak-to-peak distance function that is a function of the time (t), which is the sum of the time between the nearest peaks between the first peak column and the second peak column, is calculated. A peak-to-peak distance function calculating means;
Tempo calculation means for calculating the tempo of the arbitrary music based on the calculated period of the peak distance function;
A music tempo extraction device comprising:   A music tempo extraction program for causing a computer to realize the music tempo extraction method according to claim 1.

Images