JP2002116754A - Tempo extraction device, tempo extraction method, tempo extraction program and recording medium - Google Patents

Abstract

(57) [Summary] [Problem] Does not require a priori information for input music,
To realize a tempo extraction device that accurately extracts a tempo from a music signal. SOLUTION: A music signal is inputted to a sounding time detecting means 11 and converted into a sounding time series signal. Autocorrelation calculating means 12
Calculates the autocorrelation of the sounding time-series signal, and the peak position detecting means 13 detects the peak position of the autocorrelation pattern.
The beat structure analysis means 14 analyzes the beat structure of the music from the peak position of the autocorrelation pattern and its level. The temporary tempo calculating means 15 calculates a tempo candidate from the peak value of the autocorrelation function. The tempo determination means 16 estimates the most appropriate tempo from the beat structure and the tempo candidates. In this case, the tempo can be accurately extracted without requiring a priori knowledge of the input music.

Description

DETAILED DESCRIPTION OF THE INVENTION

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a tempo extracting device and a tempo extracting method for extracting a tempo from a music signal provided in the form of an audio signal by a CD, a broadcast or the like, a tempo extracting program, and a recording medium. is there.

[0001]

2. Description of the Related Art Tempo indicates the speed of music progression.
Often expressed in quarter notes per minute. As an example of a conventional tempo extraction device, for example,
There is a "tempo extraction device used for an automatic music transcription device or the like" disclosed in Japanese Patent No. 7751. FIG. 5 shows a block diagram of this conventional tempo extraction device. This tempo extraction device includes a signal capturing unit 31, a power calculating unit 32,
It is configured to include a measure time length calculating means 37 including a differentiating means 33, an autocorrelation calculating means 34, and a local maximum value detecting means 35, and a tempo calculating means 36. The measure time length calculation means 37 indicated by a dotted line in FIG. 5 is a block having a function for obtaining a reference measure time length.

[0002] A signal fetching means 31 samples and fetches a music signal. The power calculation means 32 calculates the power of the music signal captured for each processing frame. The differentiator 33 performs a differential calculation of the power obtained by the power calculator 32 for each processing frame. The auto-correlation calculating means 34 calculates an auto-correlation function of the differential value of the power output from the differentiating means 33. The local maximum value detecting means 35 detects the local maximum value of the autocorrelation function to determine the periodicity of the music, and calculates a reference measure time length. The tempo calculating means 36 calculates the tempo of the music from the reference measure time length and the number of beats of the music separately input.

In the measure time length calculating means 37, first, the power change of the fetched music signal is obtained by the power calculating means 32 and the differentiating means 33, and the periodicity of the power change is obtained by the autocorrelation calculating means. And the maximum value detecting means 3
In step 5, the time period in which the periodicity appears most strongly is determined based on the range of the length that a person naturally feels as one beat. That time period is the measure bar length that is the reference,
By dividing the measure time length by the number of beats, the number of quarter notes per minute, that is, the tempo is calculated.

[0004]

However, the maximum value of the autocorrelation function of the power change does not always appear in the time period corresponding to the bar time length. For example, in the case of music mainly composed of rhythm musical instruments such as recent popular music, for example, when the accent of a snare drum appears strongly in a half-note cycle, the maximum value of the autocorrelation function of the power change is two minutes. It often appears in a time period corresponding to the time length of a note period. In this case, if this local maximum value is regarded as a measure time length, there is a possibility that the maximum tempo is estimated as a double tempo. Further, in the conventional example, it is necessary to input the number of beats and the like in advance using a keyboard, and to extract the tempo requires a priori knowledge of the music to be input.

[0005] The present invention has been made in view of such conventional problems, and does not depend on the type of music to be input. An object of the present invention is to realize a tempo extraction device and a tempo extraction method that can accurately extract a tempo without requiring knowledge.

[0006]

According to the first aspect of the present invention, there is provided a sounding time detecting means for detecting a sounding time and a level of each constituent sound from an input music signal, and outputting as a sounding time series signal; Autocorrelation calculating means for calculating an autocorrelation function of a sounding time-series signal detected by the sounding time detecting means, peak position detecting means for detecting a position and a level of a peak portion from an output signal of the autocorrelation calculating means, A beat structure analyzing means for analyzing a beat structure of an input music signal from a pattern of a position and an output level of a peak portion detected by the peak position detecting means, and a position of the peak portion detected by the peak position detecting means A temporary tempo calculating means for calculating a tempo candidate of the music signal from the output signal and the output level, and a tempo candidate calculated by the temporary tempo calculating means and the beat structure solution. Based on the beat structure obtained in section, it is characterized in that it comprises a and determining tempo determining means tempo of the input music signals.

According to a second aspect of the present invention, in the tempo extraction device according to the first aspect, the beat structure analyzing means sorts the peaks detected by the peak position detecting means on the basis of their level values. Grouping means for dividing each peak rearranged by the sorting means into groups having similar level values; and parameters relating to a beat structure including the number of beat hierarchies from the level group grouped by the grouping means. And a beat structure parameter calculating means for calculating the beat structure parameter.

According to a third aspect of the present invention, in the tempo extraction device according to the first aspect, the beat structure analyzing means is obtained by a histogram generating means for generating a histogram of peak level values and the histogram generating means. Beat structure parameter calculation means for calculating parameters related to the beat structure including the number of beat layers from the histogram;
It is characterized by having.

According to a fourth aspect of the present invention, the onset time and the level of each constituent sound are detected from an input music signal,
A sounding time detecting step of outputting as a sounding time series signal, an autocorrelation calculating step of calculating an autocorrelation function of the sounding time series signal detected in the sounding time detecting step, and a peak portion from the output signal of the autocorrelation calculating step A peak position detecting step of detecting a position and a level of, and a beat structure analyzing step of analyzing a beat structure of an input music signal from a position and an output level pattern of a peak portion detected in the peak position detecting step, A tentative tempo calculating step of calculating a tempo candidate of the music signal from the position and output level of the peak portion detected in the peak position detecting step, and a tempo candidate calculated in the tentative tempo calculating step and the beat structure analyzing step. Judge the tempo of the input music signal based on the beat structure And tempo determining step, it is characterized in that it has a.

A tempo extraction program according to a fifth aspect of the present invention is characterized in that the tempo extraction method according to the fourth aspect is described in a format executable by a computer.

A recording medium according to a sixth aspect of the present invention is characterized in that the program for the tempo extraction method according to the fourth aspect is described in a format executable by a computer.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A tempo extracting device according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a tempo extraction device according to the present embodiment. This tempo extracting device includes a sounding time detecting means 11, an autocorrelation calculating means 12, a peak position detecting means 13, a beat structure analyzing means 14, a temporary tempo calculating means 1
5, including the tempo determination means 16.

A part (about 30 seconds) of a music signal provided in the form of an audio signal by a CD or broadcast is input to the tempo extraction device of the present embodiment as an input signal. From the input music signal, the sounding time detecting means 11 generates sounding start time (onset tim) of each constituent sound, for example, a sound of a snare drum, a bass drum, a bass, a guitar, a vocal, and the like.
The rising part of the sound which can be estimated to correspond to e) is detected. The sounding time detecting means 11 generates a sounding time-series signal of the input music signal based on the time and the level.

As an example of the method of detecting the onset time from a music signal, for example, Goto and Muraoka: "Beat tracking system for music sound signal-Detection of number of measures and selection of musical knowledge according to presence or absence of percussion sound-", There is a technology described in Information Processing Society of Japan, 97-MUS-21-8, Vol. 97, No. 67, pp. 45-52, 1997. Here, the power of each frequency component of the input music signal is obtained by FFT for each frame of a fixed length, and the rising portion of the sound, that is, the sounding start time of each constituent sound is estimated based on the degree of power change between frames. Then, by arranging the estimated sound generation start time and the power level of the signal at that time on the time axis,
A sounding time-series signal of the input music signal can be generated.

The autocorrelation calculating means 12 calculates an autocorrelation function of the sounding time-series signal of the music signal. X [n] is the sounding time-series signal in frame n, m is the delay time,
Assuming that the calculated time length is N frames, the autocorrelation function A [m] with respect to the delay time m is obtained by the following equation.

(Equation 1)

FIG. 2 shows an example of the autocorrelation function calculated as described above. The tempo is calculated based on the autocorrelation function.
The peak position detecting means 13 obtains a peak portion of the autocorrelation function, that is, a portion showing a maximum value. In the example of FIG. 2, the part marked by a white circle corresponds to the peak part.

The beat structure analyzing means 14 analyzes the beat structure of the input music signal from the peak portion of the autocorrelation function detected by the peak position detecting means 13. The auto-correlation function calculated by the auto-correlation calculating means 12 expresses the periodicity of the sounding portion of the input music signal. For example, if a bass drum is pronounced in a music signal and the bass drum is regularly pronounced every quarter note, a strong peak should appear at a portion corresponding to the quarter note in the autocorrelation function. is there. Therefore, by observing the peak position and the level of the autocorrelation function, the periodicity of the sounding time of each component sound of the input music, that is, the beat can be analyzed. Here, the beat structure represents a rhythm system of each constituent sound constituting the music, and specifically corresponds to a beat of the music and each note (a 16th note, an 8th note, a 4th note, a 2nd note, etc.). Represents the pronunciation frequency and intensity at the part where it is performed. FIG.
In the example shown in FIG.
It is understood from FIG. Each layer indicates the intensity of a beat corresponding to a note of a specific length (for example, a quarter note or the like).

FIG. 3 is a block diagram showing a configuration example of the beat structure analyzing means 14A. This beat structure analysis means 14
A includes a sorting unit 41, a grouping unit 42, and a beat structure parameter calculating unit 43. A procedure for analyzing the beat structure having such a configuration will be described. The sorting means 41 arranges the peak points of the autocorrelation function detected by the peak position detecting means 13 in FIG. 1 in order of the magnitude of the level. Then, peaks having similar levels are grouped. The grouping means 42 divides the data into a plurality of groups based on the peak level. Then, the beat structure parameter calculating unit 43 outputs the number of divided groups as the number of beat layers (in FIG. 2, the number of layers is 4) which is one of the parameters for explaining the beat structure.

FIG. 4 is a block diagram of a beat structure analyzing means 14B having another configuration example. The beat structure analysis means 14B includes a histogram generation means 51 and a beat structure parameter calculation means 52. FIG.
The difference from the beat structure analyzing means 14A is that a histogram generating means 51 is provided as means for grouping the peaks of the autocorrelation function. When the histogram generation means 51 generates a histogram based on the peak, the histogram shows a local maximum value near the level of the peak point at which the level is approximated. For this reason, the beat structure parameter calculating means 43 calculates the beat structure parameter using the peak of the maximum value as a measure of the distribution of the group.

The operation of the tempo extraction device having such a configuration will be described. The temporary tempo calculating means 15 in FIG. 1 calculates a tempo candidate which is considered to be the tempo of the input music from the peak portion detected by the peak position detecting means 13.
Usually, a constituent sound is often generated with an accent in units of one measure, two beats (half notes), or one beat (quarter note), and the peak portion of the autocorrelation function is often used. The tempo candidate can be estimated with the maximum value of. For example, in popular music in recent years, the sound of a snare drum is often pronounced as an accent on the second and fourth beats (that is, two beat cycles).
It can be easily expected that the peak becomes maximum in the time portion corresponding to the beat cycle.

In the example of FIG. 2, the peak 21 is the maximum value of the peak portion, and the time between the peaks is one bar, two beats, or one bar.
It is considered to represent the time length of the beat. Time of peak 21 (period of 100 frames where one frame is 86 ms)
Thus, a tempo candidate that is the number of quarter notes per minute is calculated. As a result, assuming that the peak 21 has one bar period, 207
BPM is 103 BPM for two beat cycles, and 52 BPM for one beat cycle. These three are assumed to be temporary tempos in FIG. Note that BPM (Beats Per Minute) is a unit representing the number of quarter notes in one minute.

Based on the beat structure obtained by the beat structure analyzing means 14, for example, the number of beat layers, the tempo determining means 16 determines from the temporary tempo candidates obtained by the temporary tempo calculating means 15 that it is most suitable for the input music signal. Determine possible tempo. The number of beat hierarchies is one of the important parameters for judging the tempo, and it is known from the analysis result of general popular music that a song with a fast tempo has a small number of beat hierarchies (generally, the number of hierarchies is 3 or less). ing. For example, if the number of beat hierarchies is 4 in a song having a provisional tempo of 220 BPM and 105 BPM, it can be determined that the tempo of the song is more likely to be 105 BPM. This is consistent with the rule of thumb of popular music that a deep-layered sound, that is, a sixteenth-note sound rarely appears periodically and frequently in a song with a fast tempo such as 220 BPM.

In the example of FIG. 2, a peak 22 having a double cycle is observed at a level close to the peak 21 in the hierarchy 1, and hierarchies 2 to 4 in which the levels are lowered every half cycle appear. . Thus, layer 1 is a period corresponding to a two-beat period (half note), similarly, layer 2 is a one-beat period (quarter note), layer 3 is a 0.5-beat period (eighth note), and layer 4 Is considered to represent the peak of 0.25 beat cycle (sixteenth note).

The layer 1 can be considered as one bar period. In this case, the level of the layer 2 or lower is often larger than that of the present example. Observed by type. For this reason, in the case of this example, it is considered that it is most appropriate to set the layer 1 to a two-beat cycle. Therefore, 103 BPM, which is a tempo candidate when the layer 1, that is, the peak 21 corresponds to a two-beat cycle, is output as the tempo of the input music signal.

In the present embodiment, a music signal having an autocorrelation function as shown in FIG. 2 has been described as an example, but the present invention can be applied to a music signal having another autocorrelation function pattern. it can.

[0026]

According to the tempo extraction device of the present invention, the beat structure of music is analyzed from the pattern of the autocorrelation function of the input sounding time-series signal of music, and the most significant value is obtained from the peak value of the autocorrelation function and the beat structure. By estimating the tempo that seems appropriate, the tempo can be accurately extracted without depending on the type of the input music and without requiring a priori knowledge of the input music such as the number of beats. be able to.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a tempo extraction device according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an example of an autocorrelation function obtained by a tempo extraction device according to the present embodiment.

FIG. 3 is a beat structure analysis unit 14 according to the present embodiment.
FIG. 2 is a block diagram illustrating a configuration example of A.

FIG. 4 is a beat structure analysis unit 14 according to the present embodiment.
FIG. 4 is a block diagram showing a configuration example of B.

FIG. 5 is a block diagram showing the configuration of a conventional tempo extraction device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 Onset time detecting means 12 Autocorrelation calculating means 13 Peak position detecting means 14, 14A, 14B Beat structure analyzing means 15 Temporary tempo calculating means 16 Tempo judging means 21, 22 Layer 1 peak 41 Sorting means 42 Grouping means 43 Beat structure Parameter calculation means 51 Histogram generation means

Claims (6)

[Claims] 1. An onset time detecting means for detecting an onset time and a level of each constituent sound from an input music signal and outputting the onset time as a onset time series signal, and an onset time series signal detected by the onset time detecting means Autocorrelation calculating means for calculating an autocorrelation function of; a peak position detecting means for detecting a position and a level of a peak part from an output signal of the autocorrelation calculating means; and a position of a peak part detected by the peak position detecting means. And a beat structure analyzing means for analyzing a beat structure of the input music signal from the output level pattern, and a tempo candidate of the music signal is calculated from the position and the output level of the peak portion detected by the peak position detecting means. A temporary tempo calculating means, based on a tempo candidate calculated by the temporary tempo calculating means and a beat structure obtained by the beat structure analyzing means; Tempo determining means for determining the tempo of the input music signal,
A tempo extraction device comprising: 2. The beat structure analyzing means, wherein: a sort means for rearranging the peaks detected by the peak position detecting means on the basis of the level value thereof; Grouping means for dividing into approximate groups; and a level group grouped by the grouping means,
The tempo extraction device according to claim 1, further comprising: a beat structure parameter calculating unit configured to calculate a parameter related to a beat structure including the number of beat layers. 3. The beat structure analyzing means includes: a histogram generating means for generating a histogram of peak level values; and a beat for calculating a parameter relating to a beat structure including the number of beat hierarchies from the histogram obtained by the histogram generating means. The tempo extraction device according to claim 1, further comprising: a structure parameter calculating unit. 4. An onset time detecting step of detecting the onset time and level of each constituent sound from an input music signal and outputting the onset time as a onset time series signal; and a onset time series signal detected in the onset time detection step. An autocorrelation calculating step of calculating an autocorrelation function of; a peak position detecting step of detecting a position and a level of a peak portion from an output signal of the autocorrelation calculating step; and a position of a peak portion detected in the peak position detecting step. And a beat structure analyzing step of analyzing a beat structure of the input music signal from a pattern of the output level, and a tempo candidate of the music signal is calculated from a position and an output level of the peak portion detected in the peak position detecting step. A temporary tempo calculating step; a tempo candidate calculated in the temporary tempo calculating step; Based on the beat structure obtained in the analysis step, the tempo extraction method characterized by having a tempo determination step of determining the tempo of the input music signals. 5. A tempo extraction program, wherein the tempo extraction method according to claim 4 is described in a format executable by a computer. 6. A recording medium characterized in that the program of the tempo extraction method according to claim 4 is described in a format executable by a computer.