JP2000267671A - Device and method for musical sound waveform analysis, and recording medium where musical sound waveform analyzing program is recorded - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make traceable the track of peak points which are successive along the time base of a frame by providing a peak track determining means, etc., which determines a next successive peak point among detected peak points according to analytical data on peak points predicted by a predicting means. SOLUTION: A CPU 1 performs a waveform analyzing process as to specified musical sound waveform data in a data memory 3 according to a waveform analyzing program stored in a program memory 2 and detects peak points of frequency spectra of respective frames of a Fourier analyzing process. Predicted amplitude data are found through predictive operation based upon multiple amplitude data as track data on peak points having amplitude data closest to the predicted amplitude data. Thus, the track of a next frame is predicted according to frequency data on the peak points which are already extracted as a peak track to extract peak points, so naturally connecting tracks can be traced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION The present invention provides a Fourier transform process for a musical tone waveform to be analyzed to obtain a frequency spectrum of a time-series frame, and detects a peak point of amplitude data of analysis data in each frame. The present invention relates to a musical tone waveform analyzer, a musical tone waveform analysis method, and a recording medium on which a musical tone waveform analysis program is recorded, which tracks a locus of the peak point continuous in the time axis direction of a frame.

[0002]

2. Description of the Related Art Conventionally, tone analysis and (re) synthesis (Analysis & (Re) Syn) for analyzing the properties of musical tones and synthesizing musical tone waveforms
The following processes are performed in the technical field of the thesis). First, a frame is cut out by multiplying a time-series sampling waveform by a window function, and a fast Fourier transform (FFT) is performed.
Processing is performed to obtain frequency spectrum analysis data. The analysis data of the frequency spectrum includes data on the frequency axis, that is, frequency data, amplitude data, and phase data, and detects all peak points (frequency values) that can be regarded as peaks of the amplitude data. Then, the window of the window function is moved to update the frame, the above processing is repeated, and the peak point of the amplitude data of the frequency spectrum in each frame is detected.

Here, the peak point in each frame is
Generally, a plurality of peaks are detected corresponding to a plurality of frequencies, and these peak points correspond to the fundamental tone of the original sampling waveform, harmonic components corresponding to the harmonics, noise components, side lobes of the FFT window function, and the like. Detected. However, the perceived pitch of the musical tone is mainly obtained by the fundamental tone and lower harmonic components, and the characteristics of the musical tone are analyzed and the musical tone waveform is synthesized so that the timbre can be obtained mainly by the higher harmonic components. In order to do so, it is important to extract peak points corresponding to fundamental and overtones. Since the peaks corresponding to the fundamental tone and the overtone form a long trajectory continuous in the time axis direction of the frame in the FFT, it is required to track the trajectory of the peak point in the processing of the musical tone waveform analysis.

FIG. 6 is a diagram showing an example of a peak of amplitude data of a frequency spectrum obtained as a result of analysis by FFT. The example shown in this figure is a piano C4 key (center C)
(The fundamental frequency is about 261.63 Hz), the horizontal axis is time (unit is ms), the vertical axis is frequency (unit is Hz), and each frame corresponding to the time window is The detected peak points are represented by dots. Also,
Horizontal thin lines represent the fundamental frequency and the frequency of its theoretical harmonic. In this figure, all points exist independently, and you can see the line that draws the trajectory in the figure, but the information that this trajectory is connected has not yet been obtained, and each frame Only information on which frequency is peaked for each time is obtained. It should be noted that the portion other than the portion that is viewed as a long trajectory is considered to be noise or the side lobe of the FFT window function is displayed.

In order to extract peak points forming a locus from such peak point data, generally, components having harmonic relations are picked in each frame. That is, the known fundamental frequency FB and its integral multiple,
The peak point closest to the frequency of 2 × FB, 3 × FB,... Is determined to be the fundamental tone or harmonic, and harmonic data is picked up for each frame. Then, it is determined that the peak points detected for the fundamental tone or the same harmonic between the frames form one locus.

[0006]

FIG. 7 is an enlarged view of the fundamental tone of FIG. 6 around 15 msec. In this figure, when the tracking direction of the locus of the peak point goes from right to left, it is natural that the peak point is connected in the direction of the arrow at the frame time A. However, in the above-described conventional method, for example, when looking at the fundamental frequency (in this case, 261.63 Hz), the following is performed under the condition that the peak point closest to the fundamental frequency is selected in each frame. The following problems occur. For example, in each frame from the right side of FIG. 7 to the frame time A, there is one peak near the fundamental frequency, but at the frame time A-1 following the frame time A, the peak point b is smaller than the peak point a. Occurs near the fundamental frequency. Therefore, frame times A to A-1
, The locus may be connected to the peak point b, and the locus of the peak point may not be tracked well.

SUMMARY OF THE INVENTION It is an object of the present invention to enable a locus of a peak point of each frequency spectrum in a time series frame of a musical tone waveform to be analyzed to be tracked with high accuracy in a locus of a peak point continuous in the time axis direction of the frame. I do.

[0008]

According to a first aspect of the present invention, there is provided a musical tone waveform analyzer for performing a Fourier transform process on a musical tone waveform to be analyzed to obtain a frequency spectrum of a time-series frame, and analyzing data in each frame. In the musical tone waveform analyzer that detects the peak point of the amplitude data of the frame and tracks the trajectory of the peak point continuous in the time axis direction of the frame, a plurality of peak points determined as the peak point continuous in the time axis direction of the frame Storage means for storing the analysis data of the peak point; prediction means for predicting the analysis data of the next peak point based on the analysis data of a plurality of continuous peak points already determined stored in the storage means; Peak trajectory determining means for determining the next continuous peak point from the detected peak points based on the analysis data of the peak point predicted by the predicting means , Characterized by comprising a.

According to the musical tone waveform analyzer of the first aspect, the storage means stores analysis data of a plurality of peak points determined as continuous peak points in the time axis direction of the frame (for example, Frequency data) is stored.
The predicting means predicts the analysis data of the next peak point based on the analysis data of the plurality of continuous peak points already determined stored in the storage means, and the peak trajectory determining means calculates the predicted peak point. The next continuous peak point is determined from the detected peak points based on the analysis data of the above, so that the analysis data of the already determined continuous peak point is used for determining the peak point to be a new trajectory. Reflected, the trajectory of the peak point can be tracked with high accuracy.

[0010] According to a second aspect of the present invention, there is provided a musical tone waveform analyzing method.
Fourier transform processing is performed on the musical tone waveform to be analyzed to obtain a frequency spectrum of a time-series frame, and a peak point of amplitude data of analysis data in each frame is detected.
In the musical tone waveform analysis method for tracking the trajectory of the peak point continuous in the time axis direction of the frame, when tracking the trajectory of the peak point continuous in the time axis direction of the frame, at least a plurality of continuous peak points After deciding,
Predict the analysis data of the next peak point based on the analysis data of a plurality of consecutive peak points that have already been determined, and, based on the analysis data of the predicted peak point, select the next one from among the detected peak points. A continuous peak point is determined, and the locus of the peak point is tracked.

According to the musical tone waveform analysis method of the second aspect configured as described above, in the same manner as in the first aspect, the determination of the peak point to be a new trajectory is performed by analyzing the already determined continuous peak point. Since the data is reflected, the locus of the peak point can be tracked with high accuracy.

According to the third aspect of the present invention, a tone waveform to be analyzed is subjected to Fourier transform processing to obtain a frequency spectrum of a time-series frame, and a peak point of amplitude data of analysis data in each frame is detected. And a recording medium storing a tone waveform analysis program for executing, by a computer, a process of tracking the trajectory of the peak point continuous in the time axis direction of the frame, wherein the peak is continuous in the time axis direction of the frame. When tracking the trajectory of a point, after determining at least multiple consecutive peak points,
A step of predicting the analysis data of the next peak point based on the analysis data of a plurality of continuous peak points already determined; and, based on the analysis data of the predicted peak point, from among the detected peak points. And a step of determining a next continuous peak point.
According to the execution of the musical tone waveform analysis program recorded on the recording medium, the same operation and effect as those of the first and second aspects can be obtained.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram of a musical tone waveform analyzer according to an embodiment of the present invention, showing a case where the personal computer having a CPU 1 is used.
The program memory 2 is a hard disk drive or a CD-RO
This is an M device or other external storage device, and stores a tone waveform analysis program described later. Also,
The data memory 3 is a RAM or the like, and stores tone waveform data to be analyzed, analysis results, and the like. The input device 4 and the performance operator 5 are a keyboard and the like, and the display 6 is a CRT or a liquid crystal display. The tone synthesizer 7 is a tone generator board or tone generator equipped with various LSI chips for synthesizing a tone based on the analysis result.
Further, the network interface 8 is MIDI,
An interface for connecting to various networks 9 such as a LAN and a telephone line.

The CPU 1 performs a waveform analysis process, which will be described later, on the designated tone waveform data in the data memory 3 based on a waveform analysis program stored in, for example, a hard disk device of the program memory 2, and performs each frame of the Fourier analysis process. At the peak point of the frequency spectrum. Then, a peak point forming a locus between frames is extracted, and analysis data of the peak point in each frame forming the locus is obtained as an analysis result.

The CPU 1 converts frequency data, amplitude data, and phase data, which are analysis data of the peak points forming the locus, for each frame based on the locus of the peak points obtained as a result of the waveform analysis described later. By outputting to the tone synthesizing unit 7 in accordance with the frame time, the tone synthesizing unit 7 performs Fourier synthesis. Further, a difference between the original waveform data and the Fourier-combined waveform data is obtained as residual waveform data. The waveform data obtained by performing Fourier synthesis with the residual waveform data is re-synthesized by the tone synthesizer 7, converted from digital to analog, and generated as a tone signal. In addition, desired waveform data can be obtained by modifying the analysis data and the residual waveform data output to the tone synthesizer 7.

FIG. 3 is a flowchart of a waveform analysis process executed by the CPU 1 in the embodiment. The operation of the waveform analysis process in the embodiment will be described with reference to FIG. FIG. 5 is a diagram showing an example of the time waveform of the sound of the C4 key of the same piano as that of FIG. The waveform data of this time waveform is time-series sampling data obtained by sampling and quantizing the amplitude of the musical tone waveform at, for example, a sampling frequency of 44.1 kHz, and is stored in the data memory 3.

When the waveform analysis processing is started, first, in step S1, a frame analysis of a sound waveform to be analyzed is performed. In the frame analysis process, first, in order to perform a Fourier analysis process, an analysis range, a window size, an FFT size,
Initial settings such as setting the HOP size for moving the window function along the time axis are performed. Generally, it is appropriate that the FFT window size be an integral multiple of the fundamental period of the waveform, for example, eight times. The sound of the C4 key shown in FIG.
63 Hz, and the number of sampling points included in one cycle is 168.6 points because the sampling frequency is 44.1 kHz. Therefore, the window size is set to 1348 points (rounded to the decimal point), which is eight times this.

Further, the FFT size must be a power of 2, and the larger the FFT size, the better the frequency resolution. However, if it is too large, it takes a long calculation time. Therefore, the minimum power number of 2 exceeding the window size is selected, and the number is set to 2048 points. Further, the HOP size is set to 1/8 of the basic period and 21 points. Under these conditions, a window function is moved on the time axis with respect to the waveform data of FIG. 5 to perform a Fourier analysis process on each of the extracted frames, thereby obtaining frequency spectrum analysis data for each time-series frame. . The analysis data of the frequency spectrum is frequency data, amplitude data, and phase data as data on the frequency axis.

Next, in step S2, the peak of the amplitude data of the analysis data of each frame in the frequency spectrum is extracted and, for example, a line spectrum component as shown in FIG. Which consists of the following.) Although FIG. 6 shows only the frequency data (vertical value) corresponding to each peak point, the amplitude data and the phase data corresponding to each peak point are also stored in the data memory 3 at the same time.

Next, in step S3, a base frame search process for determining a base frame serving as a base point for peak tracking is performed. In this base point frame search processing, a point in time at which the envelope of the time waveform in FIG. 5 becomes maximum (hereinafter, referred to as an attack max point) is detected. This is because there is a high possibility that all of the peak points that need to be followed are at the attack max point where the amplitude level is the maximum. In the time waveform of FIG. 5, the attack max point can be found at 42.9 ms and the sampling frequency is 44.1 kHz.
Since it is z, 42.9 ms is obtained as the 1892th point. Therefore, a frame whose center is closest to the attack max point (1892th point) is detected, and the detected frame is set as the 0th frame, and is set as the base frame of the peak tracking start. Then, peak tracking is performed toward a frame that is later (future) in time from the 0th frame, and when the upper limit of the analysis range is reached, the lower limit of the analysis range is shifted from the 0th frame to a frame that is earlier (past) in time. Perform peak tracking until reaching. As a result, it is possible to perform detection without causing a peak that needs to be followed.

FIG. 4 is a diagram showing the detection results of the peak points in the 0th frame. In FIG.
PK # 0 # 1, PK # 0 # 2, PK # 0 # 3, ..., PK # 0 # 9 are detected. Here, in this embodiment, when performing peak tracking, frequency data is used as analysis data for predicting a peak point of a next frame leading to a trajectory. Therefore, the frequency data of each peak point stored in the data memory 3 is described as follows. The frequency data of the m-th peak point counted from the low frequency side of the k-th frame is represented by PK # k # m. Therefore, assuming that the number of peak points found in the k-th frame is max (k), the frequency data of all the peak points in the k-th frame is PK # k # 1, PK #
# k # 2, PK # k # 3,..., PK # k # max (k). Note that the notation of the frequency data is appropriately used to indicate a corresponding peak point. Also, frames temporally subsequent to the 0th frame are referred to as a first frame, a second frame,..., And frames temporally preceding the 0th frame are referred to as a −1st frame, a −2nd frame, respectively. , ...

In peak tracking, PK # 0 # 1,
Generally, a plurality of trajectories appear corresponding to the respective peak points of PK # 0 # 2,..., PK # 0 # max (0), and the trajectories are tracked at each frame time. Since they are independent and perform the same processing for each trajectory, one trajectory will be mainly described below. Therefore, a certain peak point of the k-th frame obtained by the peak tracking for any one trajectory and the (k + 1) -th frame and the (k +)-th frame
The frequency data at the peak point of two frames,.
(k), PK # k + 1 # M (k + 1), PK # k + 2 # M (k + 2),... That is, M (k), M (k + 1), M (k + 2), etc., are the respective frames (k, k + 1, k +
It is the number of the peak point (extracted peak point) counted from the low frequency side in 2), and is a value obtained as a result of peak tracking. Therefore, the peak points constituting the trajectory are, for example, PK # k # 1, PK # k + 1 # 1, PK # k + 2 # 2,... (The first peak point of the k-th frame → the k + 1-th frame) (The first peak point → the second peak point of the (k + 2) th frame), the number of the peak point may be different.

Next, in step S4, the index i indicating the frame number is set to "0", and in step S7, "end of target frame range processing" is determined, and in step S8, the index i is incremented. The processing of peak point extraction in S5 and step S6 is repeated for frames within the analysis target range temporally after the 0th frame.

The processing of step S5 and step S6 is performed for each peak point PK # i + 1 # 1, PK # i + 1 # 2,..., PK # i + 1 # max of the (i + 1) th frame for each locus. This is a process of extracting a peak point leading to the trajectory up to the i-th frame from (i + 1). First, in step S5, for each trajectory, the frequency data of the peak point of the i-th frame and the past r (preset number) frames forming the same trajectory
PK # i # M (i), PK # i-1 # M (i-1), PK # i-2 # M (i-2),…, PK
From # ir # M (ir) (also referred to as “trajectory data”), the peak point position of the (i + 1) th frame, that is, frequency data (... -1 # M (i-1), frequency data PK ′ # i + 1 # M (i + 1) connected to PK # i # M (i) (hereinafter “predicted frequency data”
That. ) Is performed.

PK ′ # i + 1 # M (i + 1) = pred [corr [PK # i # M (i), PK # i−1 # M (i−1), PK # i−2 # M (i-2), ..., PK # i -r # M (ir)]] ... (1) where corr [] is PK # i # M (i), ..., PK # ir # M (ir) And pred [] is a prediction function using the elements of the set corr [] as variables. As the prediction function, for example, a linear combination formula f = A0 × PK # i # M (i) + A1 × PK # i-1 # M (i−1) +... + Ar × PK # i−
r # M (ir) (where A0, A1,..., Ar are coefficients), other polynomial approximations, and the least squares method.

In the processing of step S5, in each processing from the first 0th frame to the (r-1) th frame, the frequency data PK #
i # M (i), PK # i-1 # M (i-1), PK # i-2 # M (i-2),…, PK # ir
Since at least all of #M (ir) has not been detected, the frequency data that has not been detected is calculated assuming, for example, the same data as the frequency data of the 0th frame.

Next, in step S6, based on the predicted frequency data PK '# i + 1 # M (i + 1), each peak point PK # i + 1 # 1, PK # i + 1 # of the (i + 1) th frame. 2,..., PK # i + 1 # max (i + 1), an optimum peak point is extracted as a locus passing point. In this peak point extraction process, for example, a peak point having frequency data closest to the predicted frequency data PK ′ # i + 1 # M (i + 1) is selected, and the peak point is used as a locus peak point. And stored in the data memory 3. This peak point extraction process is also performed for each trajectory. Then, step S7
Proceed to.

FIG. 2A is a diagram conceptually showing a relationship between the predicted frequency data and the past r peak points. In the processing of the i-th frame, the peak points forming the trajectory in the i-th frame are shown. From the trajectory data (frequency data) and the trajectory data (frequency data) of the past three (r = 3 in this example) peak points, predicted frequency data (frequency data) of the (i + 1) th frame is obtained by the prediction operation. In the (i + 1) th frame, frequency data closest to the predicted frequency data is extracted as frequency data of a peak point forming a locus.

When the processing in steps S5 and S6 has been completed for each frame within the analysis range temporally subsequent to the 0th frame, the analysis data is stored in step S9. That is, in this step S9,
In step S6, the frequency data, the amplitude data, and the phase data of the peak points which are extracted as forming the trajectory and are associated with each trajectory are stored in the data memory 3 together with the data indicating the trajectory.

Next, the index i indicating the frame number is set to "0" in step S10, the "end of target frame range processing" is determined in step S13, and step S1 is performed.
By decrementing the index i in step 4, the peak point extraction processing in steps S11 and S12 is repeated for frames within the analysis target range that is temporally previous (past) from the 0th frame. The processing of steps S11 and S12 is the same as that of step S11 except that the peak tracking is performed in the temporally forward direction from the i-th frame.
5, S6. That is, first, in step S11, the i-th trajectory forming the same trajectory
From the frequency data of the peak points of the frame and r frames (r frames temporally later than the i-th frame),
A prediction operation for predicting the peak point of the (i-1) th frame is performed by the following equation (3).

PK ′ # i-1 # M (i−1) = pred [corr [PK # i # M (i), PK # i + 1 # M (i + 1), PK # i + 2 # M (i + 2),..., PK # i + r # M (i + r)]] (3) The prediction function pred [] is, for example, a linear combination expression f = A0 × PK # i # M (i) + A1 × PK # i + 1 # M (i + 1) +… + Ar × PK #
i + r # M (i + r), other polynomial approximations, least squares, etc.

In step S12, based on the predicted frequency data PK '# i-1 # M (i-1), the peak points PK # i-1 # 1 and PK # i-1 of the (i-1) th frame are used. .., PK # i-1 # max (i-1), for example, a peak point having frequency data closest to predicted frequency data PK ′ # i + 1 # M (i + 1) A locus passing point is selected as a peak point and stored in the data memory 3 in association with the locus peak point. This peak point extraction process is also performed for each locus, and the process proceeds to step S13.

The above steps S11 and S1
When the processing of step 2 is completed for each frame within the analysis range temporally before (past) from the 0th frame, in step 15, the frequency data, the amplitude data, and the phase data of the peak point associated with each locus Is stored in the data memory 3 together with the data indicating the trajectory, and the process ends.

As described above, the trajectory in the next frame is predicted and the peak point is extracted based on the frequency data of the peak point already extracted as the peak trajectory. .

In the above embodiment, the case where frequency data is used as analysis data for predicting the next peak point of the trajectory has been described. However, analysis of amplitude data and phase data is used for prediction of peak points. It may be. For example, as shown in FIG. 2B, the peak may be tracked in the space of the frame and the amplitude by obtaining the next predicted amplitude data from the amplitude data forming the trajectory. Alternatively, as shown in FIG. 2C, the peak may be tracked in the space between the frame and the phase by obtaining the next predicted phase data from the phase data forming the trajectory.

Further, a prediction operation is performed for each of the frequency data, the amplitude data and the phase data, and an optimum peak point is extracted according to a predetermined condition based on a plurality of frequency data, amplitude data and phase data. It may be.

The analysis data (PK # i # M (i), PK # i-1 # M (i-1), PK # i-2 # M (i-2), …, PK #
ir # M (ir) etc.) uses the analysis data of the peak points actually detected in the peak detection, but the analysis data used for the musical tone synthesis etc. as a result of the peak tracking processing is the analysis data of the actually detected peak points. The analysis data may be used, or the analysis data predicted in each frame may be used.

In the embodiment, the synthesis of the waveform and the generation of the tone are performed after the waveform analysis. However, the present invention relates to the analysis of the waveform. It is not limited. Further, it is not necessary to generate the residual waveform data.

In the embodiment, the case where the musical tone waveform analysis program is stored in the program memory 2 in advance has been described. For example, a tone waveform analysis program is recorded on a CD-ROM, and the tone waveform analysis program is loaded from a CD-ROM device onto a hard disk. And CPU
The system 1 develops the musical tone waveform analysis program of the hard disk in a RAM or the like, and controls the operation of the musical tone waveform analysis based on the RAM program in the same manner as in the above embodiment. This allows the CPU to perform the same operation as when the tone waveform analysis program is stored in the program memory. This makes it possible to easily install, add, or upgrade the tone waveform analysis program. Alternatively, the tone waveform analysis program may be recorded on a floppy disk, a magnetic disk (MO), or the like, and supplied to the RAM or the hard disk.

The tone waveform analysis program may be downloaded using the network interface 7. At this time, for example, a connection is made to a network 9 such as a LAN (local area network), the Internet, or a telephone line, and the distribution of the musical tone waveform analysis program from the server computer via the network 9 is recorded on the hard disk. To complete the download. Further, a tone waveform analysis program may be executed through a network.

The present invention is not limited to the personal computer as in the above embodiment, but includes various devices such as various electronic musical instruments, tone generators, sequencers, and effectors.
It may be incorporated as a function or an application in a system or the like that connects each device using communication means such as IDI or various networks.

The medium on which the musical tone waveform analysis program described in the above embodiment is recorded, ie, the RO
A destination storage device such as M, RAM, hard disk, CD-ROM, magneto-optical disk, DVD (digital versatile disk), or network server computer corresponds to a medium storing the tone waveform analysis program according to claim 3 of the present invention. I do.

[0043]

As described above, according to the first aspect of the present invention,
According to the musical tone waveform analyzer of the present invention, the musical tone waveform analyzing method of the second aspect, or the execution of the musical tone waveform analyzing program recorded on the recording medium of the third aspect, the peak point to be a new trajectory is already determined. The analysis data of the continuous peak points is reflected, and the locus of the peak points can be tracked with high accuracy.

[Brief description of the drawings]

FIG. 1 is a block diagram of a musical tone waveform analyzer according to an embodiment of the present invention.

FIG. 2 is a diagram conceptually illustrating a trajectory tracking process in the embodiment.

FIG. 3 is a flowchart of a waveform analysis process in the embodiment.

FIG. 4 is a diagram illustrating a detection result of a peak point in a 0th frame according to the embodiment.

FIG. 5 is a diagram illustrating an example of a time waveform of a sound of a C4 key of the piano according to the embodiment.

FIG. 6 is a diagram illustrating an example of a peak of a frequency spectrum according to the embodiment.

FIG. 7 is a diagram illustrating a conventional problem.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... CPU, 2 ... Program memory, 3 ... Data memory, 7 ... Tone synthesizer

Claims (3)

[Claims] 1. A frequency spectrum of a time-series frame is obtained by performing a Fourier transform process on a musical tone waveform to be analyzed, a peak point of amplitude data of analysis data in each frame is detected, and the peak point is detected in a time axis direction of the frame. A musical sound waveform analyzer that tracks a trajectory of the continuous peak points, a storage unit that stores analysis data of a plurality of peak points determined as continuous peak points in a time axis direction of the frame, and a storage unit that stores the analysis data. Prediction means for predicting the analysis data of the next peak point based on the analysis data of the plurality of consecutive peak points which have been determined, and the analysis means for detecting the peak data based on the analysis data of the peak points predicted by the prediction means. And a peak locus determining means for determining a next continuous peak point from the peak points. 2. A frequency spectrum of a time-series frame is obtained by performing a Fourier transform process on a musical tone waveform to be analyzed, a peak point of amplitude data of analysis data in each frame is detected, and the peak point is detected in a time axis direction of the frame. In the musical tone waveform analysis method for tracking the trajectory of the continuous peak point, when tracking the trajectory of the continuous peak point in the time axis direction of the frame, after determining at least a plurality of continuous peak points, it is already determined The analysis data of the next peak point is predicted based on the analysis data of a plurality of consecutive peak points, and the next continuous peak is detected from the detected peak points based on the analysis data of the predicted peak point. A tone waveform analysis method characterized in that a point is determined and a locus of a peak point is tracked. 3. A frequency spectrum of a time-series frame is obtained by performing a Fourier transform process on a musical tone waveform to be analyzed, a peak point of amplitude data of analysis data in each frame is detected, and the peak point is detected in a time axis direction of the frame. A recording medium recording a tone waveform analysis program for executing a process of tracking the trajectory of the continuous peak point by a computer, when tracking the trajectory of a continuous peak point in the time axis direction of the frame, After determining at least a plurality of continuous peak points, a step of predicting analysis data of a next peak point based on analysis data of a plurality of continuous peak points already determined, and analyzing the analysis data of the predicted peak points. Determining the next consecutive peak point from the detected peak points based on the A recording medium recording a waveform analysis program.