JPH05273964A - Attack time detection device used for automatic music transcription device etc. - Google Patents

Abstract

(57) [Summary] [Object] An object of the present invention is to provide an apparatus capable of accurately detecting an attack time by utilizing a change in spectrum. [Configuration] The CPU is sampled by an A / D conversion device
A power spectrum is calculated for the music signal data stored in the AM at regular time intervals, and time series F ₁ , F ₂ , ..., F _N of the power spectrum are calculated (S1).
Next, the CPU calculates the time series ΔF ₁ , ΔF ₂ , ..., ΔF _N of the time change amount with respect to the power spectrum (S2). Next, the CPU uses the power
The total of the positive components of the time change amount with respect to the spectrum is calculated to calculate the time series S ₁ , S ₂ , ..., _SN (S3).
Next, the CPU detects the local peaks of the time series S ₁ , S ₂ , ..., S _{N and} sets the time at which the local peaks are detected as the attack time (S4).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic music transcription device for musical notation of an acoustic music signal, and more particularly to a device for detecting the rising time (hereinafter, attack time) of each sound.

[0002]

2. Description of the Related Art Conventionally, an attack time detection device used for this kind of automatic music notation device has been constructed as shown in FIG. The audio amplifier 1 receives the played music signal as an input and amplifies this signal to an appropriate voltage value. The low-pass filter 2 suppresses aliasing distortion at the time of sampling by passing only the frequency component of 5.5 kHz or less in the signal amplified by the audio amplifier 1. The A / D converter 3 converts the low-pass filtered signal into a 16-bit digital signal with a sampling frequency of 12 kHz. I / O port 4 is connected to CPU 5 and A /
The D conversion device 3 and the display 8 are connected. CP
U5 performs a time variation calculation process of logarithmic power of music signal data, a peak detection process, etc., and a RAM 6 and a ROM.
Connected to 7.

The RAM 6 is provided with an area for storing the music signal data sampled by the A / D converter 3 and the time variation of the logarithmic power calculated by the CPU 5. The ROM 7 stores logarithmic power time change amount calculation logic, peak detection logic, and the like. The display 8 displays processing results and the like.

The operation of the conventional example will be described below with reference to FIGS.

FIG. 5 is a flow chart showing the processing performed by the conventional example.

First, the CPU 5 converts the music signal data x ₁ , x ₂ , ..., X _H (H indicates the number of data samples) sampled by the A / D converter 3 and stored in the RAM 6. On the other hand, the time series P of short-time logarithmic power is set for each frame set at 120 point (10 msec) intervals.
₁ , P ₂ , ..., P _N are calculated (S51).

The short-time logarithmic power time series calculation process S51 will be described with reference to FIG. First, the CPU
5 initializes the frame number i (hereinafter, frame number) for processing the music signal data (i = 0: S6).
1). Next, the music signal data stored in the RAM 6 is input, and the short-time power R _i is calculated by the following equation (S63).

[0008] Next, the CPU 5 calculates the logarithmic power P _i for the short-time power R _i by the following equation and stores the result in the RAM 6 (S64).

P _i = log (R _i ) Next, the CPU 5 increments the frame number i and returns the process to S62 (S65).

In S62, the CPU 5 compares the pointer 120xi indicating the observation position with the sample number H of the music signal data, and the judgment of 120xi> H is "NO". S63 to S65 described above. Processing is repeated, and if the determination is “YES”, the processing S51 ends (S6).
2).

[0011] Then, CPU 5 is determined in the processing S51, the time-series P _1, P ₂ of the logarithmic power stored in the RAM 6, · · ·, chronological time variation amount with respect to P _N delta
P ₁ , ΔP ₂ , ..., ΔP _N are calculated and stored in the RAM 6 (S52).

Next, the CPU 5 detects the local peak of the time series ΔP ₁ , ΔP ₂ , ..., ΔP _N of the amount of change in logarithmic power, and sets the time at which the local peak is detected as the attack time (S53). ).

The local peak detection processing S53 will be described with reference to FIG. First, the CPU 5 initializes the frame number i (i = 0: S41).

Next, the CPU 5 determines whether ΔP _i is a local peak. The following judgment formula is used.

ΔP _i ≧ ΔP _i−1 AND ΔP _i ≧ ΔP _{i + 1} AND ΔP _i > (certain threshold value) When the determination is “YES”, the CPU 5 executes the process S
44, if "NO", skip S44 and S4
The process proceeds to 5 (S43).

In process S44, the CPU 5 writes the frame number i as an attack time at a predetermined address on the RAM 6 (S44).

Next, the CPU 5 increments the frame number i and returns the processing to S42 (i = i + 1: S45).

In S42, the CPU 5 determines whether or not the processes of S43 to S45 have been completed for all the frames (i> N). If the determination is "YES", the entire process is completed.

FIG. 7 shows an example of the processing result. The horizontal axis of FIG. 7 represents the frame number (time axis). The upper graph plots the amplitude of the music signal sampled by the A / D converter 3 on the vertical axis. The time marked by a vertical line in the upper row is the true attack time. The middle graph is a graph in which the amount of change in logarithmic power with time calculated in step S52 and stored in the RAM 6 is plotted on the vertical axis. The time marked with a vertical line in the middle row is the attack time detected by the conventional example (in the middle row, the straight line drawn horizontally with respect to the horizontal axis shows the threshold value).

[0020]

However, in the attack time detecting device used in the above-mentioned automatic music transcription device or the like, the attack time is detected by detecting the peak of the time change amount of the short time power of the music signal. Since, for example, the sound immediately before the instrument sound to be detected is strong like the second attack time shown in the upper part of FIG. 7,
Moreover, when the duration is long, the short-time power rise at the attack time of the musical instrument sound to be detected is completely hidden, and as a result, the attack time cannot be detected accurately (the middle part of FIG. 7). ..

The present invention solves the above-mentioned problems, and an object of the present invention is to provide an apparatus capable of accurately detecting the attack time by utilizing the change in spectrum.

[0022]

In order to achieve this object, the invention according to claim 1 calculates a power spectrum for acoustic music signal data at regular time intervals and time series of the power spectrum. F ₁ , F ₂ , ..., F _N (where F _i = {F _i1 , F _i2 , ..., F _iM } is the power spectrum of the i-th frame, and M is the frequency band division number).
, And a time-series ΔF ₁ , ΔF ₂ , ..., Δ for calculating the time variation of the power spectrum.
F _N (However, ΔF _i = {ΔF _i1 , ΔF _i2 , ..., ΔF _iM }
Indicates the time variation of the power spectrum of the i-th frame. ) And a means for outputting the time series S ₁ , S ₂ , ..., _SN by calculating the sum of the positive components of the time change amount of the power spectrum for each frame by the following equation. When, However, max (a, b) comprises a, value the time series of whichever not smaller of _{b S 1, S 2, ···} , and a peak detection means for detecting the local peak of the S _N.

[0023]

The frequency analyzing means of the present invention having the above configuration calculates the power spectrum for the acoustic music signal data at regular time intervals and calculates the power spectrum time series F ₁ , F ₂ ,. , F _N are output. The differentiating means calculates the time change amount of the power spectrum to obtain a time series ΔF ₁ ,
Outputs ΔF ₂ , ..., ΔF _N. Time series S ₁ , S ₂ , ...
.., S _N calculation means calculates the sum of the positive components of the time change amount of the power spectrum for each frame by the following equation and outputs time series S ₁ , S ₂ , ..., S _N. However, max (a, b) is the smaller value of a and b. The peak detection means detects the local peaks of the time series S ₁ , S ₂ , ..., _SN , and the local peaks are detected. The output time is output as the attack time.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an automatic music transcription device according to this embodiment. The audio amplifier 1 constituting the present embodiment receives the played music signal as an input and amplifies this signal to an appropriate voltage value. Low-pass filter 2 is 5.5kHz in the signal amplified by audio amplifier 1.
The aliasing distortion at the time of sampling is suppressed by passing only the following frequency components. The A / D converter 3 passes the low-pass filter signal to a sampling frequency of 12 kHz.
Convert to z, 16-bit digital signal. The I / O port 4 connects the CPU 5, the A / D conversion device 3, and the display 8. The CPU 5 performs frequency analysis processing of music signal data, peak detection processing, and the like, and is connected to the RAM 6 and the ROM 7. The RAM 6 has an A / D
Music signal data sampled by the converter 3, CPU
An area for storing the frequency analysis result processed by the CPU 5, the attack time detected by the CPU 5, and the like is prepared. The ROM 7 stores frequency analysis logic, peak detection logic and the like. The display 8
Displays the processing result.

The operation of this embodiment will be described below with reference to FIGS.
Will be described.

FIG. 2 is a flow chart showing the processing performed by this embodiment.

First, the CPU 5 responds to the music signal data x ₁ , x ₂ , ..., X _H (H is the number of data samples) sampled by the A / D converter 3 and stored in the RAM 6. , for each frame set at 120 points (10 msec) intervals, the time series F _1, F ₂ of the short-time log power spectrum, ..., and calculates the F _N (S1).

Here, the short-time logarithmic power spectrum time series calculation process S1 will be described with reference to FIG. First, the CPU 5 initializes a frame number i (hereinafter referred to as a frame number) for processing music signal data (i =
0: S31). Next, using the music signal data stored in the RAM 6 as an input, the short-time spectrum Q _i = {Q _i1 ,
Q _i2 , ..., Q _iM : M indicates the number of frequency band divisions. }
Is calculated by the discrete Fourier transform represented by the following equation.

[0030] However, m = 1, 2, ..., M In this embodiment, M = 256 (S33).

Next, the CPU 5 calculates a logarithmic power spectrum F _i = {F _i1 , F _i2 , ..., F _iM } for the short time spectrum calculated in S33 by the following equation,
The result is stored in the RAM 6 (S34).

F _im = log (| Q _im | ² ) where m = 1, 2, ..., M Next, the CPU 5 increments the frame number i and returns the process to S32 (S35).

In S32, the CPU 5 compares the pointer 120 × i indicating the observation position with the sample number H of the music signal data, and the judgment of 120 × i> H is “NO”. S33 to S35 explained above. If the determination is “YES”, the short time logarithmic power spectrum calculation process S1 is ended (S32).

Next, the CPU 5 obtains the time-series ΔF ₁ of the time change amount with respect to the time series F ₁ , F ₂ , ..., F _N of the logarithmic power spectrum stored in the RAM 6 obtained in the process S1. , ΔF ₂ , ..., ΔF _N (where ΔF _i =
{ΔF _i1 , ΔF _i2 , ..., ΔF _iM } indicates the time change amount of the power spectrum of the i-th frame. ) Is calculated,
It is stored in the RAM 6. In this embodiment, a linear regression coefficient based on a logarithmic power spectrum for 7 frames is used as the time variation (S2).

[0035] Next, the CPU 5 obtains the RAM in the process S2
For the time series ΔF ₁ , ΔF ₂ , ..., ΔF _N of the time variation with respect to the logarithmic power spectrum stored in 6, the sum of the positive components is calculated for each frame by the following formula, and the time series S ₁ , S ₂ , ..., _SN are calculated (S3).

[0036] However, max (a, b) is a value that is not smaller than a and b. Next, the CPU 5 detects the local peaks of the time series S ₁ , S ₂ , ..., _SN , and detects the local peaks. The attacked time is defined as the attacked time (S4).

The local peak detection processing S4 will be described with reference to FIG. First, the CPU 5 initializes the frame number i (i = 0: S41).

Next, the CPU 5 determines whether or not S _i is a local peak. The following judgment formula is used.

S _i ≧ S _i−1 AND S _i ≧ S _{i + 1} AND S _i > (certain threshold value) The CPU 5 executes the process in S4 when the determination is “YES”.
If "NO", go to S4 and proceed to S45 (S43).

When the process proceeds to S44, the CPU 5 writes the frame number i as the attack time at a predetermined address on the RAM 6 (S44).

Next, the CPU 5 increments the frame number i and returns the processing to S42 (i = i + 1: S45).

In S42, the CPU 5 determines whether or not the processes of S43 to S45 have been completed for all the frames (i> N), and if the determination is "NO", S43 to S43.
If the determination is “YES” by repeating the process of S45, all the processes are finished (S42).

FIG. 7 shows an example of the processing result. The horizontal axis of FIG. 7 represents the frame number (time axis). The upper graph plots the amplitude of the music signal sampled by the A / D converter 3 on the vertical axis. The time marked by a vertical line in the upper row is the true attack time. The lower graph shows the time series S ₁ , S ₂ , ..., S calculated in the process S3 and stored in the RAM 6.
_N is plotted on the vertical axis. It can be seen that the time marked with a vertical line in the lower row is the attack time detected by this embodiment, and is also detected for the second attack time (in the lower row, horizontally with respect to the horizontal axis). The drawn straight line indicates the threshold value).

The present invention is not limited to the configuration described in detail above, and various modifications can be made without departing from the spirit of the invention.

[0045]

As is apparent from the above description, according to the attack time detection device used in the automatic music transcription device or the like of the present invention, the power spectrum is obtained at regular time intervals with respect to the acoustic music signal data. Frequency analysis means for calculating and outputting time series F ₁ , F ₂ , ..., F _N of the power spectrum, and time series ΔF ₁ , ΔF ₂ , ... For calculating the time change amount of the power spectrum. , ΔF _N and a sum of positive components of the time variation of the power spectrum is calculated for each frame by the following equation to obtain a time series S ₁ ,
Means for outputting S ₂ , ..., _SN , However, max (a, b) has a, value the time series of whichever not smaller of _{b S 1, S 2, ···} , and a peak detection means for detecting the local peak of the S _N, the local By setting the time at which the peak is detected as the attack time, it is possible to detect the attack time with high accuracy.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of the present invention.

FIG. 2 is a flowchart showing the operation of the present invention.

FIG. 3 is a flowchart showing the operation of the frequency analysis processing of the present invention.

FIG. 4 is a flowchart showing an operation of peak detection processing.

FIG. 5 is a flowchart showing an operation of a conventional example.

FIG. 6 is a flowchart showing the operation of a conventional logarithmic power calculation process.

FIG. 7 is an explanatory diagram showing an attack time detection processing result.

[Explanation of symbols]

 1 Audio amplifier 2 Low pass filter 3 A / D converter 5 CPU 8 Display

Claims (1)

[Claims] 1. A power spectrum for acoustic music signal data is calculated at regular time intervals, and time series of power spectrum F ₁ , F ₂ , ..., F _N (where F _i = {F _i1 ，
F _i2 , ..., F _iM } indicates the power spectrum of the i-th frame, and M indicates the number of frequency band divisions. ), And a time series ΔF ₁ , ΔF ₂ , ..., ΔF _N (however, ΔF _i
= {ΔF _i1 , ΔF _i2 , ..., ΔF _iM } indicates the time variation of the power spectrum of the i-th frame. ) And a means for outputting the time series S ₁ , S ₂ , ..., _SN by calculating the sum of the positive components of the time change amount of the power spectrum for each frame by the following equation. When, However, max (a, b) has a, value the time series of whichever not smaller of _{b S 1, S 2, ···} , and a peak detection means for detecting the local peak of the S _N, the local An attack time detection device used in an automatic music transcription device or the like, wherein an attack time is a time at which a peak is detected.