JP2003263162A - Method and device for estimating tempo of musical data - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To accurately estimate a tempo even when processing musical data of which the tempo is changed on the way. <P>SOLUTION: In [analysis 1], a tempo is estimated by detecting beat positions over the whole input audio data. In [integration 1], analytical results are integrated to obtain a tempo list. Since there are sections for which analysis is failed, detailed analysis for these sections is performed in [analysis 2]. In [integration 2], the analytical results of individual sections are integrated to be collected as the tempo information of the whole input audio data. Since there are analyzed sections having the same result but divided into pieces or an extremely short center section of which analytical result is different from right and left sections, a section for which an analytical section is judged to be extended is analyzed again in [analysis 3]. In [integration 3], these analytical results are compared/integrated to be collected as the tempo information of the whole input audio data. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a music data tempo estimation method, and more particularly to a method and apparatus for accurately estimating the tempo even for music data whose tempo changes during a song.

[0002]

2. Description of the Related Art Conventionally, there is a technique for detecting the beat (and estimating the tempo) of audio data of a phrase having a beat feeling by regarding the tempo as constant. For example, JP-A-62-205397, JP-A-63-8697,
JP-A-63-70293, JP-A-63-70294
JP-A-5-27751, JP-A-7-64544
No. 2000-221979 and the like describe an apparatus for extracting tempo from an acoustic signal.

[0003]

All of the above-mentioned conventional techniques detect a beat from a phrase having a constant tempo and estimate the tempo. Therefore, it cannot be applied to the tempo estimation of music data in which the tempo changes in the middle of a song. Even if it is forcedly applied, there is a problem that the position where the tempo changes cannot be accurately recognized, and as a result, the tempo cannot be accurately estimated.

In view of the above-mentioned problems in the prior art, the present invention provides a method for processing music data that enables accurate estimation of the tempo even when processing music data whose tempo changes midway. An object is to provide a tempo estimation method and device.

[0005]

In order to achieve this object, the invention according to claim 1 is a method for estimating a tempo of music data, in which a step of inputting music data to be estimated and a step of inputting music data are performed. A step of setting a plurality of different estimation ranges; a step of analyzing the music data for each of the estimation ranges to estimate a tempo to obtain a tempo value for each of the estimation ranges; and a tempo value of each of the estimation ranges. Integrating to obtain a temporally varying tempo value of the music data.

According to a second aspect of the present invention, in the music data tempo estimating method according to the first aspect, the estimation range is a plurality of estimation ranges having different widths.

The invention according to claim 3 is the music data tempo estimating method according to claim 1, wherein the estimation range is a plurality of estimation ranges having different positions.

According to a fourth aspect of the present invention, in a music data tempo estimation method, a step of inputting music data to be estimated, a step of setting a plurality of different estimation ranges for the input music data, and the estimation Analyzing the music data for each range to estimate the tempo, obtaining a tempo value and a score value indicating the reliability of the estimation for each estimated range; Integrating based on the score value to obtain a temporally varying tempo value of the music data.

According to a fifth aspect of the present invention, in the music data tempo estimating method according to any one of the first to fourth aspects, the integration of the tempo values of the plurality of estimation ranges is extracted from the music data. It is characterized in that the integration is performed on the basis of the determined beat position.

According to a sixth aspect of the present invention, in the method for estimating the tempo of music data, the step of inputting the music data to be estimated and the two estimated ranges on the partially overlapped music data are estimated ranges. Analyzing the music data of to estimate the tempo and obtain the tempo value,
With respect to the overlapping portion, modifying at least one estimation range of the two estimation ranges so that the two estimation ranges do not overlap, and integrating tempo values of the two estimation ranges at least one of which is modified. And obtaining a temporally varying tempo value of the music data.

According to a seventh aspect of the present invention, in the method for estimating the tempo of music data, the step of inputting the music data to be estimated and the two estimation ranges to be continuously joined are respectively set to the music within the estimation range. Analyzing the data to estimate the tempo and obtain the tempo value;
Modifying the tempo value or the beat position of at least one of the two estimated ranges so that the beat positions of the two estimated ranges do not deviate from each other at the junction; and the tempo value of the two estimated ranges with at least one modified Are integrated to obtain a tempo value that changes with time of the music data.

According to an eighth aspect of the present invention, in a method for estimating tempo of music data, a step of inputting music data to be estimated and a plurality of ranges on the time axis of the music data are respectively included in the music of the estimated range. The step of analyzing the data to estimate the tempo to obtain the tempo value; and, of the plurality of ranges, the distance between adjacent ranges is less than or equal to a predetermined width, and the difference between the tempo values estimated in each range is A step of grouping a plurality of ranges having a predetermined value or less and determining an expected tempo of each group, and for a range in which each group exists, music data in the range is analyzed to estimate the tempo and the tempo value is determined. The step of obtaining and the tempo value obtained in each group and the expected tempo of each group are compared. If the difference is less than a predetermined value, the Integrating range, characterized in that it comprises a step of the integrated tempo values obtained in the group tempo value in the range.

According to a ninth aspect of the present invention, in the method for estimating tempo of music data, a step of inputting music data to be estimated and a plurality of first frames having a first width are set at different positions on the music data. Then, estimating the tempo for each first frame at each position, and setting a plurality of second frames having a second width smaller than the first width at different positions on the music data, Estimating a tempo for every two frames;
A step of integrating a tempo value estimated in a frame and a tempo value estimated in the plurality of second frames to obtain a fluctuating tempo value of the music data.

According to a tenth aspect of the present invention, in the method for estimating the tempo of music data, the step of inputting the music data to be estimated, the music data is divided into predetermined sections, and the expected range of tempo values is set for each section. The steps to set
Analyzing the music data to estimate the tempo value of each section, and for each section, leave the estimated tempo value within the expected range and delete the estimated tempo value that does not fall within the expected range. And a step.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows the overall configuration of a waveform tempo estimation system according to an embodiment of the present invention. This system includes a central processing unit (CPU) 101, a read only memory (ROM) 102, and a random access memory (RA).
M) 103, drive device 104, MIDI interface 106, panel switch 107, panel display 1
08, a writing circuit 111, an access management unit 112, a waveform memory 113, a sound source 114, a sound system 115, and a bus line 120.

The CPU 101 is a processing device that controls the overall operation of this system. ROM 102 is a CPU
The nonvolatile memory 101 stores various control programs executed by 101 and constant data. RAM103
Is a volatile memory used for a load area and a work area of a program executed by the CPU 101. The drive device 104 is a CD-ROM or a hard disk (H
D) or the like for connecting the external storage device 105.
The MIDI interface 106 is an interface for connecting to an external MIDI device. The panel switch 107 is various operators provided on the external panel of this system. The panel display 108 is a display for displaying various information provided on an external panel of this system.

The writing circuit 111 is a circuit for writing the waveform data input from the outside into the waveform memory 113. The access management unit 112 uses the write circuit 111, C
Waveform memory 113 from PU 101 and sound source 114
Control for arbitrating access requests to the. The waveform memory 113 is a storage device that stores waveform data input from the outside. The sound source 114 accesses the waveform memory 113 according to an instruction from the CPU 101, reads predetermined waveform data, and generates a musical tone signal. Sound system 1
15 emits sound based on the tone signal output from the sound source 114.

The system of FIG. 1 may be configured as an electronic musical instrument having a keyboard or the like, or may be configured by connecting a tone generator board or the like to a general-purpose personal computer.

FIG. 2 shows an outline of an operation procedure for estimating the tempo of the musical tone waveform data in the system of FIG. In step 201, audio data (waveform data) for which the tempo is estimated is prepared. The audio data may be selected from waveform data input (recorded) from the outside by a user operating a recording switch or the like to give an instruction, or waveform data stored in the external storage device 105. The audio data to be processed is RAM1
The data is stored in the work area on 03 and processed, but the data on the waveform memory 113 may be directly processed. In step 202, the upper limit value and the lower limit value of the estimated tempo are input by operating the operator. Since the user often recognizes a tempo range that is about this level for the audio data to be processed, the user is prompted to enter the range. One tempo value range may be specified for the entire audio data to be processed, or the tempo value range may be specified for each section by dividing into sections.
In step 203, the tempo estimation button is operated to instruct the start of tempo estimation processing. The tempo estimation process will be described in detail later. In step 204, the result of tempo estimation is displayed on the panel display 108. After confirming the result, an operation of saving the result of tempo estimation is performed, and the result of tempo estimation is received.

Next, the tempo estimation processing executed in step 203 will be described in detail.

FIG. 3 is a diagram showing a procedure of tempo estimation processing and a data flow. The tempo estimation processing is executed by performing the processing in the order of [analysis 1] → [integration 1] → [analysis 2] → [integration 2] → [analysis 3] → [integration 3].

The outline of each process will be described. In [Analysis 1], while moving a frame (frame for cutting out data to be processed along the time axis in the input audio data) over the entire input audio data, the data in each frame is moved from the beginning to the end of the data. The beat position detection processing is applied to. By the beat position detection processing, the beat position within the frame can be detected. Analysis result 1 is a simple collection of these beat position detection results. In [Integration 1], the analysis results of the respective sections of the analysis result 1 are integrated and summarized as tempo information of the entire input audio data. Since there are sections where the analysis does not work well, sections that are judged to be finely analyzed again are collected and passed to [Analysis 2] as section list 1. The data used in the subsequent integration process is passed to [Integration 2] as the intermediate result 1.

In [Analysis 2], the beat position detection process is applied to the data in each frame while moving the frame for each of the intervals included in the interval list 1.
Compared with [Analysis 1], the width of the frame and the moving step are reduced. Analysis result 2 is a simple collection of these beat position detection results. In [Integration 2], the analysis results of each section of the intermediate result 1 and the analysis result 2 are integrated and summarized as tempo information of the entire input audio data. There are some analysis sections that have the same result but the analysis section is fragmented, or only the very short middle section has different results from the left and right, so the sections that are judged to be wider should be collected. Pass it to [Analysis 3] as List 2. The data used in the subsequent integration process is passed to [Integration 3] as the intermediate result 2.

In [Analysis 3], the beat position detection processing is directly performed on each of the sections included in the section list 2 without cutting out the data with a frame. Analysis result 3 is a simple collection of these beat position detection results. In [Integration 3], the analysis results of the respective sections of the intermediate result 2 and the analysis result 3 are compared / integrated and summarized as tempo information of the entire input audio data. The result is the final result. This is the final output of the system of this embodiment. Specifically, this output is obtained in the form of a tempo list showing its constituent units in FIG. This data will be described in detail later.

Next, each process will be described in detail.

First, the beat position detection process will be described. The beat position detection processing is performed by [analysis 1], [analysis 2],
In [Analysis 3] and [Analysis 3], the processing is performed on the set analysis range (frame). By the beat position detection processing, the beat position within the analysis range can be detected.

FIG. 4 shows an example of the procedure of beat position detection processing. In step 401, the audio data in the frame is filtered to extract a required band (frequency band component effective for detecting a beat). Step 40
In 2, the volume envelope is extracted from the audio data of the extracted band. In step 403, the volume envelope is differentiated. In step 404, the position where the differentiated volume envelope becomes larger than a predetermined threshold value BT1 and its peak value are detected. In step 405, a list BPL listing the positions and peak values detected in step 404 is created.

FIG. 17A shows the list BPL obtained by the beat position detection processing. Note that, here, various data are represented by a notation according to how to write a declaration statement of a structure of the programming language C. For example, in FIG.
7 (a), {beat position BP1, peak value PV
1} indicates set data of the data of the beat position BP1 and the data of the peak value PV1. This is "beat position BP
It indicates that there is a peak of the peak value indicated by "peak value PV1" at the position indicated by "1". The list BPL is such a beat position BP and the peak value PV at that position.
It is a list of pair data with. The arrangement order is the direction along the time axis of the input audio data (that is, the ascending order of beat positions).

Next, the processes of [Analysis 1], [Integration 1], [Analysis 2], [Integration 2], [Analysis 3], and [Integration 3] will be described in detail.

[Analysis 1] will be described. Figure 5
A procedure of beat detection processing performed in [Analysis 1] will be described.

In step 501, a frame which is the first analysis range of the input audio data to be processed is prepared. Next, in step 502, the beat position detection processing described in FIG. 4 is performed on the audio data in the frame to create the beat position list UBPL. The data format of the beat position list UBPL is shown in FIG.
It is the same as the list BPL in (a). Step 503
Then, the time interval between each beat is calculated, the time interval is divided into groups, and each group votes. This is a process of grouping the time intervals between beats according to the value and counting how many time intervals appear. In step 504, the tempo value UTV of the frame is calculated based on the peak position and the peak interval of the voting result.
To decide.

Next, at step 505, the peak level and the variance of the peak intervals of the voting result are obtained, and the score value USV is calculated from the level and the variance. The score value is an index of the reliability of the estimated reliability, that is, an index indicating the degree of success / failure of beat detection, and if the peak level of the voting result is large (at this time, the beat position is the beat position). The score value is large (because it is more sure that the indentation has been made), and the score value is smaller if the variance of the peak intervals is large (at this time, it is recognized that the time intervals between the peaks are different).

In step 506, a frame which is the next analysis range is prepared. This is a process of advancing the frame by a predetermined amount from the current frame. In step 507, it is determined whether or not the next frame is ready, and when it is not ready (that is, when the analysis is completed up to the end of the input audio data to be processed), the process is ended. If there is a next frame, the process returns to step 502, and the beat detection process is continued with the next frame as the current frame.

FIG. 17B shows the format of the data obtained as a result of performing the processing of steps 502-505 on one frame. The start point USP and the end point UEP indicate the position of this frame. That is, it is shown that the following data was obtained as a result of analysis from the start point USP to the end point UEP in the input audio data. The beat position list UBPL is calculated in step 50.
2 is a list created in FIG. 2 and is data in which beat positions in the frame and peak values at the positions are arranged (see FIG. 17).
(A) format). The score value USV is the score value calculated in step 505. Tempo value UT
V is the tempo value calculated in step 505. Note that the beat position is expressed as a relative position (offset) from the beginning of the frame, which is the analysis start point, in the analysis of FIG. 4, but in the list UBPL of FIG. It shall be converted.

FIG. 6 shows how the beat detection processing of [Analysis 1] in FIG. 5 is performed while advancing the frame. widt
h1 is the width of the frame, and hop1 is the movement distance when advancing from one frame to the next frame. In the figure,
Although about 4 to 5 beats are drawn to be included in one frame, in practice, it is preferable that the width 1 of the frame has a value such that 10 to 20 or more beats are detected. For example, the length (corresponding to 8 [sec]) of four phrases of the tempo of 120 [bpm] is set to the width.
1, and half the length may be set as hop1 and the like.
[Bpm] is a unit of tempo expressed by how many quarter notes are cut in one minute.

First, a frame 602 indicated by a dotted line is set for the input audio data 601, and steps 502 to 505 are performed in this range to obtain unit data UD having the format shown in FIG. 17B. Next, this frame is advanced by the amount indicated by hop1 to set the next frame 603 indicated by the solid line. This frame 603 is also subjected to the processing of steps 502 to 505 in the same manner as in FIG.
The unit data UD of the format shown in (b) is obtained. By repeating this, the unit data UD shown in FIG.
To get Data obtained (analysis result AR1 (AR: Analy
sis Result)) is data corresponding to the analysis result 1 in FIG.

[Integration 1] will be described below. In [Integration 1], integration of each analysis result (that is, unit UD of beat detection results) of the analysis result AR1 obtained in [Analysis 1],
Prepare for the next [Analysis 2]. The feature of this integration method is, in short, that the higher the score value, the longer the section that adopts the analysis result (for the range where the two data UD ranges overlap, the data UD with the higher score value is given priority. The integrated position of the analysis result coincides with any of the beat positions (the start point and the end point of each data UD are adjusted to always coincide with the beat position).

FIG. 7 is a flow chart showing the procedure of the integration process performed in [Integration 1].

In step 701, the data UD that does not meet the conditions is removed from the analysis result AR1. As a result, the analysis result unit UD whose score value USV is less than or equal to a certain threshold value UT1 or whose tempo is out of the set range is excluded from the analysis result AR1, and is excluded from the processing target of the next step and thereafter. The tempo setting range means step 20 in FIG.
It is the range of the tempo input by the user in 2. Step 7
In 02, the start point USP and the end point UEP of each analysis result unit UD are corrected to the beat position. This correction process is a process of adjusting the start point USP and the end point UEP of each data UD so that they coincide with the beat position. Specifically, all the analysis result units UD remaining after the removal in step 701 are converted into the unit HUD of the end point corrected detection result shown in FIG. 17C, respectively.

The unit HUD of the end point corrected detection result of FIG. 17C is the unit UD of the beat detection result of FIG. 17B.
Is created by copying, but the start point and the end point are converted into the first beat position and the last beat position. First beat position U
The SP fetches and sets the first element (first beat position) from the beat position list UBPL of the unit UD of the beat detection result of FIG. The final beat position is shown in Figure 1.
The last element (last beat position) is taken out from the beat position list UBPL of the unit UD of the beat detection result of 7 (b) and set. The beat position list UBPL may be copied as it is without changing. As a result, FIG.
In the unit data HUD of (c), data in which the delimiter of the analysis section is matched with the position of the beat is obtained. Score value US
The V and the tempo value UTV may be copied as they are.

Next, in step 703, the unit HUDs of all the obtained end-point corrected detection results are calculated as the score value USV.
In ascending order (from the smallest score value to the largest score value). In step 704, the tempo list TL1
To create. The creation of the tempo list is a process of overwriting and copying the contents of the unit HUD (FIG. 17C) of the detection result after the end point correction to the tempo list TL in order from the smallest score value. As a result, data having a higher score value is overwritten later and is more preferentially reflected in the tempo estimation result.

FIG. 8 shows the procedure of the tempo list creation processing performed in step 704 in [Integration 1]. First, in step 801, a new tempo list TL is created.
The tempo list TL is composed of a sequence of tempo list constituent units TUD.

FIG. 17D shows the tempo list constituent unit T.
The UD format is shown. Left TUSP and right TUE
P indicates a position in the input audio data. The list of beat positions TUBPL is from the left end TUSP to the right end TUE.
List of beat detection results in the range up to P (Fig. 17
(A) format). The tempo value TUTV is
It is the tempo value estimated in the range from the left end TUSP to the right end TUEP. The tempo list TL created in step 801 is the tempo list constituent unit TUD which is its element.
There is no empty list.

Next, in step 802, the unit data HUD of the end point corrected detection results that have been sorted (FIG. 17C).
The data TUD (FIG. 17 (d)) of the tempo list constituent unit is created from the first data HUD of. Specifically, the first beat position USP is at the left end TUSP, the final beat position UEP is at the right end TUEP, and the beat position list UB
PL to beat position list TUBPL, tempo value UTV
To the tempo value TUTV. Since the scores are sorted in ascending order of the USV, the first data of the sorted data UD has the smallest score USV. The tempo list constituent unit TUD created in step 802 is data to be added (copied) to the first tempo list TL.

In step 803, the following <case1> to
It is determined which of <case4> is applicable. The following is a detailed description of what each <case> is and how to process at that time.

<Case1> is the copy destination tempo list T
On the area where all the ranges (the range from the left end TUSP to the right end TUEP) covered by the existing constituent units T of L are combined, the tempo list constituent unit TUD created in step 802 to be added now from the left end TUSP to the right end TU.
This is the case where the range up to EP is empty. In the tempo list creation process of FIG. 8, basically, the sorted data H
The UDs are sequentially taken out, and the tempo list constituent unit TUD based on the sorted data HUD is added to the tempo list TL. Since the tempo list constituent unit TUD is data that defines the tempo value in the range from the left end TUSP to the right end TUEP in the input audio data, the process proceeds and the tempo list constituent unit TU in the tempo list TL is processed.
As D increases, the range in which the tempo is defined in the audio data increases. <Case1> is
Tempo list unit TUD to be newly added
Is the data that defines the tempo value for the first time in the range from the left end TUSP to the right end TUEP, in other words, in the tempo list constituent unit TUD that has been added to the tempo list TL, it is about to be added. This is a case where there is nothing that overlaps the range from the left end TUSP to the right end TUEP of the tempo list constituent unit TUD.

If it is determined as <case1> in step 803, the process proceeds to step 808 through step 807. In step 808, the tempo list constituent unit TUD to be added is pasted to the tempo list TL. On this occasion,
Left-end TUSP of tempo list unit TUD to be added
Is the rightmost TU of the tempo list unit TUD next to it
When it matches with EP, the right end T of the constituent unit TUD on the left side
The beat at the position of UEP (that is, the last beat in the range) is deleted from the beat position list TUBPL of the constituent unit TUD on the left side. Also, when the right end TUEP of the tempo list constituent unit TUD to be added matches the left end TUSP of the tempo list constituent unit TUD to the right of it, the beat at the position of the right end TUEP of the constituent unit TUD to be added is added. It is deleted from the beat position list TUBPL of the target structural unit TUD. Step 807 is
A correction process (hereinafter referred to as an end point process) is performed when it overlaps with the left or right adjacent structural unit TUD.

When the tempo list constituent unit TUD is pasted on the tempo list TL in step 808, it is added by being inserted in an appropriate position. The tempo list TL is an array of a plurality of tempo list constituent units TUD, and these tempo list constituent units TUD are sorted in ascending order with the value of the leftmost TUSP as the first key and the value of the rightmost TUEP as the second key. There is. Therefore, it is assumed that the tempo list constituent unit TUD to be newly attached is inserted at an appropriate position in this sort order.

Next, in step 809, the next unit data HUD of the sorted end point corrected detection results (FIG. 17).
(C)) is taken out, and from the unit data HUD, the next tempo list constituent unit TUD to be added (FIG. 17).
(D)) is created. In step 810, it is determined whether or not there is the next addition target data TUD. If there is a next addition target, the process returns to step 803, and if not, the tempo list creation process ends.

FIG. 9 shows the states of the tempo list before copying and the tempo list after copying in the case of <case 1>. The horizontal axis direction represents the time direction of audio data (the same applies to FIGS. 10 to 13). Reference numerals 901 and 902 respectively denote tempo list constituent units TUD already pasted in the tempo list TL (state before pasting in step 808). The length of the rectangle indicates the range of the tempo list constituent unit TUD, that is, the range from the left end TUSP to the right end TUEP of the tempo list constituent unit TUD. Now, the tempo list composition unit TUD to be added is
In the case of the data having the range from the left end to the right end shown in 903, there is no tempo list constituent unit TUD in that range of the tempo list TL, so the case is <case 1>. In this case, as shown after copying, the tempo list constituent unit TUD to be added may be pasted on the tempo list as it is. However, the data subjected to the end point processing (correction processing in step 807) for the left end and the right end of the tempo list constituent unit TUD to be added is pasted.

Next, <case 2> will be described. <Case2
> Is a case where the range of the tempo list constituent unit TUD to be added completely includes the existing tempo list constituent unit TUD on the tempo list TL. That is, in the constituent unit TUD of the current tempo list TL, there is already a constituent unit TUD completely included in the range from the left end TUSP to the right end TUEP of the tempo list constituent unit TUD to be added now. This is the case.

If it is determined as <case 2> in step 803, existing data TUD included in the pasting range of the tempo list constituent unit TUD to be added is deleted in step 804, and the process proceeds to step 807. Step 8
The processing after 07 is as described in <case1>.

FIG. 10 shows the states of the tempo list before copying and the tempo list after copying in the case of <case 2>. Reference numerals 1001, 1002, and 1003 denote tempo list constituent units TUD already pasted in the tempo list TL (state before pasting in step 808).
Indicates. Tempo list unit TUD to be added now
Is data having a range from the left end to the right end shown in 1004, 1002 is included in the range of the tempo list TL.
Since the tempo list unit shown in is included, <
case2>. In this case, as shown after copying, the data of 1002 is deleted, and the tempo list constituent unit TUD to be added is pasted to the tempo list as shown at 1004. However, the data subjected to the end point processing (correction processing in step 807) for the left end and the right end of the tempo list constituent unit TUD to be added is pasted.

Next, <case 3> will be described. <Case3
> Is a case where the range of the tempo list constituent unit TUD to be added overlaps with the range of the existing tempo list constituent unit TUD on the tempo list TL, but is not in a relationship of being completely included or included. That is, in the case where the constituent unit TUD of the current tempo list TL already has a constituent unit TUD which partially overlaps the range from the left end TUSP to the right end TUEP of the tempo list constituent unit TUD to be added now. Is.

If it is determined as <case3> in step 803, in step 805 the existing constituent unit data TU
D is converted to data TUD excluding the pasting range of the tempo list constituent unit TUD to be added. This is because the data after the score value is high is prioritized in the partially overlapping range, so that the existing tempo list unit TUD
To exclude the overlapping range. Step 80
After 5, the process proceeds to step 807. The processing after step 807 is as described in <case1>.

FIG. 11A shows the states of the tempo list before copying and the tempo list after copying in the case of <case 3>. Especially, the tempo list constituent unit TU to be added
The case where the left side of the range D and the right side of the existing tempo list constituent unit TUD overlap each other is shown. 1101,110
Reference numerals 2 and 1103 respectively denote tempo list constituent units TUD already pasted in the tempo list TL (state before pasting in step 808). If the tempo list constituent unit TUD to be added is the data in the range 1104, the tempo list constituent unit 1102 exists in that range of the tempo list TL and partially overlaps. case3>. In this case, as shown in the figure after copying, the data of 1102 is converted by deleting the overlapping portion, and the tempo list constituent unit TUD to be added is pasted to the tempo list as indicated by 1104. The tempo list constituent unit TUD to be added is subjected to the end point processing in step 807 and pasted.

FIG. 11B is a diagram showing beats deleted from the existing tempo list constituent unit TUD in the case shown in FIG. 11A. The left end position 1111 to the right end position 1112 indicate the range of the tempo list constituent unit 1102 in FIG. That is, the leftmost TUSP of this tempo list constituent unit 1102 is at position 1111,
The right edge TUEP is position 1112. As shown in FIG. 11A, when the range of the tempo list constituent unit 1102 and the range of the tempo list constituent unit 1104 to be added overlap, the right end of the tempo list constituent unit 1102 is corrected. Reference numeral 1113 in FIG. 11B indicates the corrected right end position. The beat on the right side is deleted from the tempo list constituent unit 1102.

FIG. 12A shows another tempo list before copying and another tempo list after copying in the case of <case 3>. In particular, the right side of the range of the tempo list constituent unit TUD to be added and the existing tempo list constituent unit TU
The case where the left side of the range of D overlaps is shown. 1201,12
02 and 1203 respectively indicate tempo list constituent units TUD already pasted in the tempo list TL (state before pasting in step 808). If the tempo list constituent unit TUD to be added is the data in the range 1204, the tempo list constituent unit 1202 exists in that range of the tempo list TL and partially overlaps. case3>. In this case, as shown after copying, the data of 1202 is converted in a manner that the overlapping portion is deleted, and the tempo list constituent unit TUD to be added is pasted to the tempo list as indicated by 1204. The tempo list constituent unit TUD to be added is subjected to the end point processing in step 807 and pasted.

FIG. 12B is a diagram showing beats deleted from the existing tempo list constituent unit TUD in the case shown in FIG. 12A. The left end position 1211 to the right end position 1212 indicate the range of the tempo list constituent unit 1202 in FIG. That is, the leftmost TUSP of this tempo list constituent unit 1202 is at position 1211,
The right edge TUEP is position 1212. As shown in FIG. 12A, when the range of the tempo list constituent unit 1202 and the range of the tempo list constituent unit 1204 to be added overlap, the left end of the tempo list constituent unit 1202 is corrected. 1213 of FIG.12 (b) shows the position of the left end after correction. The beat on the left side is deleted from the tempo list constituent unit 1202.

Next, <case 4> will be described. <Case4
> Is a case where the range of the tempo list constituent unit TUD to be added is completely included in the range of the existing tempo list constituent unit TUD on the tempo list TL. That is, in the case where the constituent unit TUD of the tempo list TL at the present time already has a constituent unit TUD that completely includes the range from the left end TUSP to the right end TUEP of the tempo list constituent unit TUD to be added now. Is.

If it is determined as <case4> in step 803, in step 806 the existing constituent unit data TU
D is two data TUD before and after the pasting range
Convert to. In the overlapping range, the latter data having a high score value is prioritized, and therefore, the overlapping range is removed from the existing tempo list constituent unit TUD, and the front side data and the rear side data are left. Step 806
After that, the process proceeds to step 807. The processing after step 807 is as described in <case1>.

FIG. 13 shows the states of the tempo list before copying and the tempo list after copying in the case of <case 4>. Reference numerals 1301, 1302 and 1303 denote tempo list constituent units TUD already pasted in the tempo list TL (state before pasting in step 808).
Indicates. Tempo list unit TUD to be added now
Is the data in the range 1304, the tempo list constituent unit 1302 that completely includes the range already exists, and therefore the case is <case 4>. In this case, as shown after copying, the data of 1302 is divided into two parts, the front side and the rear side, and the tempo list constituent unit TUD to be added is pasted in the meantime as indicated by 1004. However, the data subjected to the end point processing (correction processing in step 807) for the left end and the right end of the tempo list constituent unit TUD to be added is pasted.

The tempo list constituent unit TUD defines the tempo value of a certain range in the audio data, but since the tempo value is a calculated value obtained as a result of analyzing the data of the range, the range is reversed. When considering a beat that engraves an ideal beat according to the tempo value, the ideal beat position and the actually detected beat position may deviate from each other. Therefore, when the two tempo list constituent units TUD are joined (without a blank area) in each of the above-mentioned cases, the joining position is adjusted to the actual beat existing position, so that the joining position is ideal according to the tempo value. There is a possibility that it will be out of alignment with the beat position. In order to prevent the tempo from collapsing, the tempo value TUTV is set in both or one of the two tempo list constituent units TUD to be joined.
Need to be corrected appropriately. Specifically, it is easy to correct the tempo value TUTV of the front TUD of the two tempo list constituent units TUD. Tempo value TUTV from the starting point of the tempo list constituent unit TUD in the front
Then, an ideal beat is chopped, and the value of the tempo value TUTV is increased / decreased so that some beats thereof coincide with the start point of the rear tempo list unit TUD. That is, an integer number of beats must be included in each unit TUD.

In this embodiment, data is copied from the unit data HUD of FIG. 17C by the integration processing of FIG. 7 and the tempo list unit data TU of FIG. 17D is copied.
Although D is created, the unit data HUD may be directly incorporated into the tempo list TL as the tempo list constituent unit TUD. In this case, the unit data HUD is corrected according to each of the above cases. By doing so, the correction result in [integration 1] is directly carried over to [integration 2], so the processing amount in [integration 2] is reduced. However, the detection result 1 is lost.

Returning to FIG. 7 again, when the creation of the tempo list TL in step 704 is completed as described above, in step 705, a section list 1 which is a list of sections to be analyzed next is created. This is an empty section {left end, right end} from the tempo list TL1 created in step 704.
It is a process that collects only items and creates a list. This empty section is a section in which the score value is below a certain value and the tempo value cannot be acquired in the processing of [Analysis 1]. This is a section list SL1 (SL: S
section list).

FIG. 14 shows an example of data in the section list SL1. The shaded area in the tempo list of FIG. 14A (for example, the range of positions p1 to p2, the position of p3 to
The range of p4, the range of positions p4 to p5, the range of positions p6 to p7, etc.) indicates the range in which the tempo list constituent unit TUD exists and the tempo value of the range is acquired. A blank range between them is a range in which the tempo value is not acquired, and this range is registered in the section list SL1. 14
(B) shows the data format of the section list SL1. The section list SL1 is data in which the set data of the start position and the end position of the range in which the tempo value is not acquired are arranged as a list.

Returning again to FIG. 7, next step 706.
Then, prepare for the next integration process. This is a list created by copying data from the unit HUD (FIG. 17C) of the output result of step 702, that is, the end point corrected detection result. This list is the intermediate result IR1
(IR: Interim Result).

FIG. 18A shows the data format of the unit data IR1UD which is an element constituting the intermediate result IR1. In the unit data of FIG. 17C and the unit data of FIG. 18A, corresponding data elements are represented by the same symbols. For example, the first beat position USP and the starting point US
P and so on.

This completes the processing of [Integration 1]. By [Integration 1], the section list SL1 (FIG. 14) that is a list of ranges to be analyzed next, and the unit HUD of the end point corrected detection result that stores the result obtained by the analysis (FIG. 17).
Intermediate result IR1 obtained by copying (c)) (FIG. 18 (a))
To get a list of. In addition, a tempo list TL (the element of which is shown in FIG. 17D), which is a list of ranges in which the tempo values can be acquired in the input audio data, is obtained. The "section list 1" in FIG. 3 is the section list SL1, and the "intermediate result 1" is the intermediate result IR1.

[Analysis 2] will be described below. In [Analysis 2], beat detection processing is performed again for each section designated in the section list SL1 obtained in [Integration 1]. The procedure of beat detection processing is the same as that described with reference to FIG. However, in [Analysis 1], the analysis was performed while shifting predetermined frames little by little from the beginning to the end of the input audio data, but in [Analysis 2], each section of the section list SL1 is targeted, and from the beginning of the first section. The predetermined frame is gradually shifted to the end and analyzed, and then the predetermined frame is gradually shifted from the beginning to the end of the second section, and so on. The width of each section to be analyzed may be slightly widened.

The width w of the frame used in [Analysis 2]
The idth2 and the frame moving step hop2 are wi
Any value may be used as long as dth2 <width1 and hop2 <hop1, but normally width2 = width
It is sufficient to set th1 / 2 and hop2 = hop1 / 2. For example, width2 = 1 bar, hop2 = 1/4
Set as a measure. In addition, the threshold BT2 for beat extraction is BT1
Can be the same as. In [Analysis 2], a narrower frame width width2 and movement step hop2 are used than in [Analysis 1], so that a more rigorous analysis can be performed.

Similar to [Analysis 1], the analysis result AR2 is a list of units of beat detection results. Figure 1
8B shows the data format of the unit data AR2UD which is a component of the analysis result AR2. The data element of this unit data AR2UD is the same as the data element of the unit data UD of FIG. 17 (b). However, the unit data AR2UD represents the result of analyzing each section of the section list SL1. “Analysis result 2” in FIG. 3 is the analysis result AR2.

[Integration 2] will be described below. In [Integration 2], integration of each analysis result and preparation for [Analysis 3] are performed. The integration method is the same as in FIG. 7 described in [Integration 1]. However, the preparation for [Analysis 3] is significantly different from [Integration 1].

[Integration 2] according to the flowchart of FIG.
The processing procedure will be described. In step 701, data for which beat detection has failed is eliminated. This processing is similar to the processing of the same step in [Integration 1], and the analysis result AR2
From FIG. 18 (b), it is a process of excluding an analysis result in which the score value USV is less than or equal to a certain threshold value or the tempo value UTV is outside the set range. The threshold UT2 is
The value may be the same as or different from the value UT1 used in [Integration 1].

At step 702, the end point of the analysis result is corrected to the beat position. This process is also similar to the same process in [Integration 1]. As a result, each analysis result is shown in FIG.
Is converted into the unit data HUD of the detection result after the end point correction. In step 703, the result is the score value U
Sort based on SV. The result and the intermediate result 1
(The result obtained in [Integration 1], the structural unit of which is shown in FIG. 18A) are merged, and all the merged structural unit data are arranged in ascending order using the score value USV as a key. . In step 704, the tempo list TL2 is created based on the sorted data. This process is also similar to the same process in [Integration 1]. Tempo list T
The data format of L2 is the same as that shown in FIG. .

Next, at step 705, a section list SL2 which is a list of sections to be analyzed next is created. This is a process of finding all the sequences of tempo list constituent units TUD that satisfy both of the following conditions 1 and 2 from the tempo list TL2 in order to search for the sections that have been over-analyzed in [analysis 1] and [analysis 2]. is there. A series is a set of two or more tempo list constituent units UD, all of which are continuous in the sense of the elements of the list. In terms of area, there may be a space between the constituent units.

Condition 1 (condition of width of free area): The width between all adjacent tempo list constituent units TUD (in the meaning of "list element") included in a series is a certain threshold value WT (WT: Width Threshold) or less. The width between adjacent tempo list constituent units TUD means the right end TUEP of the first tempo list constituent unit TUD and the second tempo list constituent unit T on the right side thereof.
TUSP-TUEP is calculated from the left end TUSP of UD.

Condition 2 (condition of constant tempo): The difference between the maximum value and the minimum value of the tempo values of the tempo list constituent unit TUD included in the series is a certain allowable error ET (ET: Error Th).
reshold) or less. The value of the allowable error ET may be determined according to the tempo. For example, tempo 60 [bpm]
Permissible error around 2 [bpm], tempo 180 [bpm]
The permissible error may be around 6 [bpm].

The basis of these conditions is that it is highly likely that adjacent intervals are estimated to have the same tempo and form the same phrase, and that a sufficiently small interval sandwiched between intervals estimated to have the same tempo happens to occur. This means that there is a possibility that the section may be part of the same phrase as the left and right sections only if the analysis fails for that section.

FIG. 15A shows an example in which the above conditions 1 and 2 are satisfied and the conditions are not satisfied in the process of step 705. u1, u2, u3, ... Correspond to the tempo list constituent unit UD, respectively, and the length of the rectangle is the range of the tempo list constituent unit UD (from the left end TUSP to the right end TUEP).
Range). The number in the rectangle indicates the tempo value TUTV of the tempo list constituent unit UD. For example, u
1 indicates that this range is the tempo 120. Condition 1
The threshold value WT of the width is set to the width shown in 1501, and the allowable error ET of the condition 2 is 2 [bpm].

Although u1 and u2 are adjacent to each other, the width between them is larger than the threshold value WT1501. Therefore, since Condition 1 is not satisfied, it cannot be included in the section list SL2.

For u3, u4, and u5, the width between u3 and u4 and the width between u4 and u5 are both threshold WT150.
Since it is 1 or less, Condition 1 is satisfied. Also, u3, u4
The tempo value of u5 is 102, 100, 101, and the difference between the maximum value and the minimum value thereof is 2, which is equal to or less than the allowable error ET, so that the condition 2 is satisfied. Therefore, from the left end of u3, u
The rightmost section of 5 is put in the section list SL2.

U6 and u7 have a width WT between them as a threshold value WT.
Since it is 1501 or less, Condition 1 is satisfied. However, the tempo value is 100 and 80, the difference is 20 and the allowable error ET
Condition 2 is not satisfied because it exceeds, and section list SL2
I can't put it in. Similarly, tempo list unit TU
Find a series of D. Here, u10 and u11
Is recognized as a continuous range, and u13, u14 and u1
5 is recognized as a continuous range.

FIG. 15B shows a section list SL2 created by performing the processing of step 705 on the tempo list including the respective constituent elements as shown in FIG. 15A. The format of the section list SL2 is the same as that of the section list SL1 described in FIG.

Returning to FIG. 7 again, when the section list SL2 described above is created in step 705, step 706
Then, as a preparation for the next integration, the intermediate result 2 is created. First, the “expected tempo value” is added to each section that is an element of the section list SL2 to create the check section list 2. The expected tempo value is a value at which a tempo value of this range is expected to be about this level. This is added for the purpose of checking when an unexpected analysis result is obtained in [Analysis 3].
As the expected tempo value, an intermediate value between the maximum value and the minimum value under the condition 2 of step 705, and the same step 70
A plurality of tempo list constituent units U determined to be continuous in 5
It may be set to the average value of the tempo value of D or the like.

FIG. 18C shows a check section list 2
2 shows the format of the constituent unit data CSL2UD.
This unit data is data obtained by adding the data of the expected tempo value of the section to the left end and right end data of each section which is an element of the section list SL2. For example, in FIG.
The checking section list 2 created from the contents of the section list SL2 shown in FIG. 5 (b) becomes data as shown in FIG. 18 (d).

The checking section list CSL2 and the tempo list TL2 output from step 704 are intermediate results IR2. The intermediate result IR2 corresponds to the intermediate result 2 in FIG.

[Analysis 3] will be described below. In [Analysis 3], beat detection processing is performed again for each section designated in the section list SL2 obtained in [Integration 2] (the same applies when using the check section list CSL2). The procedure of beat detection processing is the same as that described with reference to FIG. However, unlike [Analysis 1] and [Analysis 2], each section specified in the section list SL2 is directly subjected to analysis without being cut out in a frame. Each section specified in the section list SL2 is a series of sections that satisfy the above conditions 1 and 2, and one tempo value can be determined in the section. Here, the threshold value BT3 for beat extraction may be the same as BT1 or BT2. [Analysis 1] and [Analysis 2]
Similar to the above, the analysis result AR3 is a list of the units of the beat detection result. The data format of the analysis result AR3 is the same as that of the analysis result AR2 described in FIG. Hereinafter, the structural unit of the analysis result AR3 is AR3.
Represented by UD. “Analysis result 3” in FIG. 3 is the analysis result AR3.

[Integration 3] will be described below. In [Integration 3], the final result is obtained by comparing the analysis result AR3 and the intermediate result IR2. In particular, the data of the analysis result AR3 is adopted as the final result in the section where the analysis result AR3 can estimate the expected tempo. Further, in the section where the analysis result AR3 cannot estimate the expected tempo, the data of the tempo list TL2 of the intermediate result IR2 is adopted as the final result.

FIG. 16 is a flowchart showing the processing procedure of [integration 3].

In step 1601, the section in which beat detection has failed is excluded from the analysis result AR3. This is the same as described in step 701 of [Integration 1] and [Integration 2], in which the analysis result AR3 indicates that the score value USV is less than or equal to a certain threshold value UT3, or that the tempo is outside the set range. This is the process of eliminating the result. Threshold UT3
May be a value different from the above-mentioned thresholds UT1 and UT2.

Next, at step 1602, the analysis result AR3
Therefore, the section where the expected tempo cannot be estimated is excluded. This is because the analysis result AR3 in which the section in which the beat detection has failed in step 1601 is excluded is compared with the check section list CSL2 of the intermediate result IR2 for each corresponding section, and the tempo value UTV of the constituent unit of the analysis result AR3 is obtained. This is a process of excluding an analysis result in which the tempo value expected of the constituent unit of the checking section list CSL2 is different. That is, the analysis result of a certain section (the constituent unit of the beat detection result) is the analysis start point USP of the constituent unit AR3UD of the analysis result AR3 = the left end of the constituent unit CSL2UD of the checking section list CSL2, and the constituent unit AR3UD of the analysis result AR3. Analysis end point UEP = tempo value UTV of the structural unit AR3UD of the analysis result AR3 with respect to the structural unit CSL2 structural unit of the check that is at the right end of the structural unit CSL2UD of the structural check list CSL2
And the unit CSL2U of the check section list CSL2
It is checked whether or not the absolute value of the difference of D from the "expected tempo value" is within the range of the allowable error TE (TE: Tolerance Error), and if it is not within the range, the structural unit of the analysis result AR3. Delete AR3UD. This allowable error TE may be set to, for example, half the allowable error of the condition 2 of step 705.

Next, at step 1603, the tempo list FR as the final result is created. This processing is the same as that described in step 704 of FIG. 7 and FIG. However, in [Integrated 1] and [Integrated 2], a new tempo list TL is created and the unit data T is added to the tempo list TL.
Although UD is overwritten, in the processing of FIG. 8 performed in step 1603, “tempo list TL2 of intermediate result IR2” is used instead of “new tempo list TL”,
The tempo list TL2 is overwritten with the tempo list constituent unit TUD converted from each constituent unit AR3UD of the analysis result AR3 after the processing of step 1602. Therefore, in step 801 of FIG. 8, instead of the “new tempo list TL”, the “tempo list TL of the intermediate result IR2 is used.
2 ”is prepared, and in steps 802 and 809, a tempo list constituent unit TUD to be added next is created from each constituent unit AR3UD of the analysis result AR3. As is clear from the processing of FIG. 8, a plurality of sections will always be overwritten.

The structural unit AR3 of the analysis result AR3
The processing of steps 802 and 809 for creating the tempo list structural unit TUD from the UD is performed as follows. Here, the format of the structural unit AR3UD of the analysis result AR3 is the same as that of the analytical result 2 of FIG. 18B, and the format of the tempo list structural unit TUD is FIG.
This is as shown in (d). Tempo list unit TU
D is created by setting the analysis start point USP of the structural unit AR3UD of the analysis result AR3 as the left end TUSP to the right end TUE.
The analysis end point UEP of the constituent unit AR3UD of the analysis result AR3 is represented as P, the beat position list UBPL of the constituent unit AR3UD of the analysis result AR3 is represented as its beat position list TUBPL, and the constituent unit AR3UD of the analysis result AR3 is represented as its tempo value TUTV. It suffices to copy each of the tempo values UTV of.

Through the above processing, the completed tempo list TL3 is the final output FR (FR: Final Result). The format of this tempo list TL3 is shown in FIG.
As shown in (d), the tempo value of each section is estimated.

According to the above embodiment, the tempo of each section can be accurately estimated even when the tempo of audio data changes in the middle of a song. Using this, for example M
The tempo of IDI data can be matched with the tempo of audio data. That is, the tempo change event may be inserted into the MIDI data at the timing when the tempo of the audio data changes by performing the processing of the above-described embodiment on the audio data. MI
For the insertion position on the DI data, the position on the MIDI data corresponding to the transition of the tempo of the audio data is calculated, and the tempo change is inserted at that position. Of the tempo change events originally present in the MIDI data, the tempo change events included in the audio data reproduction section may be deleted.

On the contrary, the tempo of the audio data can be matched with the tempo of the MIDI data by using the method of the above embodiment. The tempo of the audio data is calculated by performing the processing of the above embodiment on the audio data, and the calculated result and the tempo information of the MIDI data are compared,
It suffices to subject the audio data to time stretch processing at different scales for each section so as to match the tempo of the MIDI data. As a method of time stretching, a method of rearranging each slice piece when audio data is sliced at a beat position may be used.

It should be noted that, in the audio data whose tempo is estimated by the method of the above-described embodiment, it is relatively easy to calculate the beat position when reproducing with changing the tempo. This is because the tempo estimation processing is performed based on the idea that "tempo changes occur at beat positions." That is,
In the conventional method, since the tempo is estimated without depending on the idea that the tempo changes at the beat position, it may be estimated that the tempo changes at a position outside the beat position. In this case, if the audio data is to be reproduced with a different tempo, it is necessary to calculate both the position of the change point of the tempo and the reproduction start position of the beat position after the change of the tempo. On the other hand, in the estimation method of the above-described embodiment, the beat position coincides with the tempo change point. Therefore, when the tempo change point is reproduced, if the position of the tempo change point is calculated, the beat position is calculated. This is done, and has the merit that the amount of calculation is small.

In the above embodiment, the beat position detection processing (FIG. 4) is individually performed for each of the set analysis ranges in [Analysis 1], [Analysis 2], and [Analysis 3]. Alternatively, the beat position detection process may be performed on the entire input audio data prior to [Analysis 1], [Analysis 2], and [Analysis 3]. That is less waste of processing. Also, the beat detection positions can be made the same (or easily corresponded) between the analysis frames, which is convenient for synthesizing the analysis results of the frames later. In this case, in the processing of steps 404 and 405 of FIG. 4, the peak value is detected using a plurality of different threshold values, and a list corresponding to each threshold value is created. In [Analysis 1], [Analysis 2], and [Analysis 3], the processes may be performed using the lists to be used in the respective processes. However, if the purpose is to set the waveform processing parameters in the beat position detection processing for each analysis frame, the beat position detection processing is performed individually for each set analysis range.

In the above-mentioned embodiment, the final output tempo list also includes beat information (list of beat positions) as shown in FIG. 17D. This is because it is assumed that audio data will be sliced using this result, but it is not always necessary to include beat information in the output.

In the above embodiment, the tempo list TL3
Is the final output, but the tempo list TL1 or T
L2 may be the final output.

In the above embodiment, tempo estimation is performed on audio data (waveform data).
The tempo estimation target is not necessarily limited to audio data. For example, the present invention is applied to estimate the tempo of performance data from performance data (data represented by various event data such as a note-on event) that records a performance performed regardless of the tempo clock. You can also In this case, the beat position detecting process of FIG. 4 is a process of analyzing performance data and detecting a beat position.

In the above embodiment, when each beat is detected,
Although a frame with a constant width is used, beat detection may be performed with a frame with a width that adaptively changes according to the waveform data.

In the above embodiment, the tempo estimation is first performed with a long frame width, and the tempo estimation is performed with a short frame width for the failed range.
There is no need to do it in stages. The tempo may be estimated only once with one frame width, or the tempo may be estimated in three or more stages with three or more frame widths.

The range in which the second tempo estimation is performed does not necessarily have to be the range in which the first tempo has failed.
For example, (1) the second tempo estimation is performed including the range of the structural unit HUD (score value of which is less than or equal to a predetermined threshold value) having a low score value among the adopted ones, (2) the second "tempo""Estimation" is also performed for the entire range. (3) In the part where the two constituent unit HUDs are connected, the second tempo estimation is also performed for the range of the two unit HUDs that required a large correction of the connection position. You may

The position data such as the start point, the end point, and the beat position may be in the unit of time, or may be in the unit of the number of samples (may include a fractional part or less). The tempo value may also be a value including decimals. When the tempo value does not include a decimal point, the tempo value of each constituent unit data TUD in the tempo list may be corrected so that the error in the beat position due to the rounding of the decimal part is not accumulated.

In the above embodiment, the constituent unit data HU
Although D is sorted in ascending order of scores to create the tempo list TL, conversely, unit data HUD may be sorted in descending order of scores to create the tempo list TL. In that case, in the tempo list creation processing, the unit data TUD pasted to the tempo list TL may be prioritized, and the unit data TUD pasted later may be modified so as not to overlap it.

It is also possible to create a tempo list without performing sorting. In that case, a plurality of structural units H
The UDs are taken out in an arbitrary order and attached to the tempo list TL. Then, when the structural unit HUD to be attached later overlaps with the structural unit HUD already attached, the one with the higher score is given priority, and the structural unit H with the lower score is given priority.
The structural unit TUD created from the UD may be modified.

In the above-mentioned embodiment, the processing is carried out in the procedure shown in FIG. 3, but the re-analysis is performed on the section for which the tempo value was not obtained in [Analysis 2] (ie [Analysis 1] and [Integration 1] ) May be omitted. In that case,
For the results of [Analysis 1] and [Integration 1], the section list S is searched for an overly detailed section as described in [Integration 2].
The process of creating L2 may be performed.

Further, in the above-described embodiment, despite the two-step analysis of [Analysis 1] and [Analysis 2], it is determined in step 701 in [Integration 2] that the threshold value is not more than the threshold value, and the tempo value cannot be acquired in some sections. It may remain. Further analysis may be performed for that section. Alternatively, for a section for which the tempo could not be acquired, interpolation may be performed based on the tempo and beat position of the preceding and succeeding sections to determine the tempo of the section.

As an embodiment of the present invention, the following tempo estimating device can be realized. The first is a music data tempo estimating device, in which a frame of a predetermined width for cutting out partial data to be analyzed is set on music data whose tempo is to be estimated, and music data cut out by the frame is set. A first analysis means for obtaining a first analysis result, which is a list of tempo values for each frame, by repeating the process of calculating the tempo value while shifting the position of the frame by analyzing A means for creating a first tempo list, which is a list in which a range for which a tempo value can be obtained and a tempo value in the range are arranged from the result of the first analysis, and a tempo value in the first analysis Means for creating a first section list, which is a list in which the sections for which the detection of the first section is unsuccessful is arranged, and again, for each section of the first section list, a width narrower than that in the first analysis Second analysis means for performing a process of obtaining a tempo value using a new frame, and obtaining a second analysis result which is a list in which tempo values of those sections are arranged, and the first analysis result and the second analysis result. Means for creating a second tempo list, which is a list in which the range in which the tempo value can be obtained from the analysis result and the tempo value of the range are arranged, and a plurality of adjacent sections from the second tempo list. The distance between adjacent sections is equal to or less than a predetermined threshold, and the interval between the maximum and minimum tempo values of those sections is equal to or less than a predetermined threshold. Means for creating a second section list, which is a list of a series of sections, and a process for determining the tempo value again without setting a frame for each section of the second section list. Give and those Third analysis means for acquiring a third analysis result, which is a list in which tempo values are arranged, and tempo values for each section acquired by the third analysis, with respect to the second tempo list. It is characterized by comprising means for acquiring a final tempo list, which is a list in which a range having the same tempo value in the music data and a tempo value of the range are arranged by being reflected.

The second is a music data tempo estimating device, which sets a frame of a predetermined width for cutting out partial data to be analyzed on music data of which tempo is to be estimated, and cuts out by the frame. The first analysis result, which is a list of tempo values for each frame, is obtained by analyzing the music data to obtain the tempo value and repeating the process of obtaining the tempo value while shifting the position of the frame. Analyzing means, means for creating a tempo list, which is a list in which the range in which the tempo value can be obtained from the analysis result and the tempo value of the range are arranged, and from the tempo list, in a plurality of adjacent sections And a distance between adjacent sections is a predetermined threshold value or less, and a difference between the maximum and minimum tempo values of those sections is a predetermined threshold value or less Collectively as a section of a series, performing means for creating a segment list is a list of those series of sections, each section of the segment list, without setting the frame, again, the process of obtaining the tempo value,
Second analysis means for acquiring a second analysis result which is a list in which tempo values of the sections are arranged, and the tempo value of each section acquired by the second analysis is reflected on the tempo list. By doing so, it is characterized by comprising means for acquiring a final tempo list, which is a list in which a range having the same tempo value in the music data and the tempo value of the range are arranged.

Thirdly, in the above-mentioned music data tempo estimating device, when the tempo value is obtained, a score value as an index of how reliable the tempo value is is obtained, and the score value is obtained from the analysis result. When there are analysis results having different tempo values in overlapping sections when creating tempo lists, the tempo value having the higher score value is prioritized and reflected in the tempo list.

Fourth, in the above-mentioned music data tempo estimating device, when the tempo list is created from the analysis result, the start position and the end position of the range in which the tempo value can be obtained on the music data. It further comprises means for adjusting the range so that the position matches the beat position.

Fifth, the above-mentioned music data tempo estimation apparatus further comprises means for inputting an expected range of tempo values within a predetermined section on the music data by a user.
When the tempo value of the section outside the expected range is detected, the detected tempo value is excluded and excluded from the subsequent processing targets.

[0117]

As described above, according to the present invention,
Even for music data whose tempo changes midway, the tempo can be accurately estimated. In particular, since tempo estimation is performed in different sections and a plurality of estimation results are combined, the tempo estimated in each section is accurate. Further, since the sections in which similar tempo values are detected are integrated, it is possible to correctly detect tempos over a wide range.
Furthermore, since the tempo is estimated with the first frame length and the estimation failure period is estimated with the second frame length which is shorter than the first frame length, it is smaller than when the whole is analyzed with a short frame length. Accurate estimation can be performed with the amount of calculation. In addition, since the expected range is specified in advance for each position range, it is possible to prevent the tempo value from being estimated with no unexpected time. Since the timing of the tempo change is matched with the timing of the beat during the estimation, the processing after the estimation of the tempo (for example, the processing for reproducing another audio signal / MIDI signal in synchronization with the beat) becomes simple. If the timing of tempo change is matched with the beat timing, the correction process becomes simple when combining the results of tempo estimation performed in different sections.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of a waveform tempo estimation system according to an embodiment of the present invention.

FIG. 2 is a diagram showing a user's operation procedure when estimating the tempo of waveform data.

FIG. 3 is a diagram showing a procedure of tempo estimation processing according to the embodiment and a data flow of the entire processing.

FIG. 4 is a diagram illustrating an example of a procedure of beat position detection processing.

5 is a diagram showing a procedure of beat detection processing in analysis 1, 2, and 3. FIG.

FIG. 6 is a diagram showing how beat detection processing is performed while advancing a frame.

FIG. 7 is a diagram showing a procedure of integration processing in integrations 1 and 2.

FIG. 8 is a diagram showing a procedure of tempo list creation processing in integrations 1, 2, and 3.

FIG. 9 is a diagram showing a state of a tempo list before copying and a tempo list after copying in the case of <case 1>.

FIG. 10 is a diagram showing a state of a tempo list before copying and a tempo list after copying in the case of <case 2>.

FIG. 11 is a diagram showing a state of a tempo list before copying and a tempo list after copying in the case of <case 3>.

FIG. 12 is a diagram showing the states of another tempo list before copying and a tempo list after copying in the case of <case 3>.

FIG. 13 is a diagram showing a state of a tempo list before copying and a tempo list after copying in the case of <case 4>.

FIG. 14 is a diagram showing an example of data of a section list SL1.

FIG. 15 is a diagram illustrating an example in which conditions 1 and 2 are satisfied and a case in which conditions are not satisfied, and a data example of a section list SL2.

FIG. 16 is a diagram showing a processing procedure of integration 3;

FIG. 17 is a diagram showing formats of various data (No. 1) used in the present embodiment.

FIG. 18 is a diagram showing formats of various data (No. 2) used in the present embodiment.

[Explanation of symbols]

101 ... Central processing unit (CPU), 102 ... Read only memory (ROM), 103 ... Random access memory (RAM), 104 ... Drive device, 106 ... MIDI
Interface, 107 ... Panel switch, 108 ...
Panel display, 111 ... Writing circuit, 112 ... Access control section, 113 ... Waveform memory, 114 ... Sound source, 115 ... Sound system, 120 ... Bus line.

Claims (10)

[Claims] 1. A method for estimating a tempo of music data, the step of inputting music data to be estimated, the step of setting a plurality of different estimation ranges for the input music data, and the music data for each estimation range. And estimating a tempo to obtain a tempo value for each estimated range; and integrating each tempo value of the plurality of estimated ranges to obtain a temporally varying tempo value of the music data. A method for estimating tempo of music data, comprising: 2. The tempo estimation method for music data according to claim 1, wherein the estimation range is a plurality of estimation ranges having different widths. 3. The tempo estimation method for music data according to claim 1, wherein the estimation range is a plurality of estimation ranges whose positions are different from each other. 4. A method for estimating tempo of music data, the step of inputting music data to be estimated, the step of setting a plurality of different estimation ranges for the input music data, and the music data for each estimation range. To estimate the tempo and obtain a tempo value and a score value indicating the reliability of the estimation for each estimation range, and integrating each tempo value of the plurality of estimation ranges based on each score value. And a step of obtaining a temporally varying tempo value of the music data. 5. The music data tempo estimating method according to claim 1, wherein the tempo values of the plurality of estimation ranges are integrated based on a beat position extracted from the music data. A method for estimating tempo of music data, characterized by being integrated into. 6. A method for estimating tempo of music data, wherein the step of inputting music data to be estimated and the two estimation ranges on the music data that partially overlap each other are analyzed respectively. Estimating the tempo to obtain a tempo value, modifying at least one of the two estimation ranges so that the two estimation ranges do not overlap with respect to the overlapping portion, and at least one is modified. And a tempo estimation method for music data, which comprises: integrating tempo values of the two estimation ranges to obtain a tempo value that temporally changes of the music data. 7. A method for estimating a tempo of music data, wherein the step of inputting music data to be estimated and the two estimated ranges to be continuously joined are analyzed to determine the tempo. To obtain a tempo value, and to correct the tempo value or the beat position of at least one of the two estimated ranges so that the beat positions of the two estimated ranges do not shift from each other at the joint, Integrating the modified tempo values of the two estimated ranges to obtain a temporally varying tempo value of the music data. 8. A method of estimating tempo of music data, which comprises inputting music data to be estimated, and analyzing the tempo by analyzing the music data of each of a plurality of ranges on the time axis of the music data. And obtaining a tempo value, the distance between adjacent ranges of the plurality of ranges is less than or equal to a predetermined width, and the difference between the tempo values estimated in each of the ranges is less than or equal to a predetermined value. Group multiple ranges,
Determining the expected tempo of each group, for the range in which each group exists, analyzing the music data of the range to estimate the tempo to obtain a tempo value, and the tempo value obtained in each group Compare with the expected tempo of each group, if the difference is less than a predetermined value,
And a step of integrating a plurality of ranges belonging to the group and setting a tempo value of the combined range as a tempo value obtained in the group. 9. A method for estimating tempo of music data, the step of inputting music data to be estimated, and a first width having a first width at a different position on the music data.
Setting a plurality of frames, estimating the tempo for each first frame at each position, and setting a plurality of second frames having a second width smaller than the first width at different positions on the music data, Estimating a tempo for each second frame at each position, integrating the tempo values estimated in the plurality of first frames, and the tempo values estimated in the plurality of second frames,
And a step of obtaining a varying tempo value of the music data. 10. A method for estimating tempo of music data, which comprises inputting music data to be estimated, dividing the music data into predetermined sections, and setting an expected range of tempo value for each section. Analyzing music data to estimate the tempo value of each section; and for each section, leave the estimated tempo value within the expected range and delete the estimated tempo value outside the expected range A tempo estimation method for music data, comprising: