JPH0627940A - Automatic music transcription device - Google Patents

Abstract

PURPOSE:To obtain the automatic music transcription device which can automatically music-transcript drum sounds from multi-played music. CONSTITUTION:The automatic music transcription device is equipped with FIR (finite section impulse response) filter 21 which emphasize characteristics of respective drum sounds, power calculators 22a and 22b which calculate time series of short-time power at constant intervals of time to the outputs of the FIR filters 21a and 21b, and local peak detectors 23a and 23b which detect local peaks in the time series of short-time power; and the times and peak values of the local peaks are used as the stroke times and intensities of drums corresponding to the characteristics of the FIR filters 21.

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 recording a drum sound from a music piece that has been multi-played such as a CD.

[0002]

2. Description of the Related Art Conventionally, as an automatic music transcription device, for example,
It was published as an intelligent transcription system in the Journal of the Japan Society for Artificial Intelligence published in January 990. This intelligent transcription system was able to perform only the transcription process for musical instruments playing a scale.

[0003]

However, in the conventional automatic transcription device, since it is not possible to transcrib the drum sound from a music piece that is overplayed such as a CD, knowledge and experience are required when transcribing the drum sound. There was a drawback in that a specialist with a computer had to listen to it and write it down.

The present invention has been made in order to solve the above-mentioned problems, and it is an object of the present invention to provide an automatic transcription device capable of transcribing a drum sound from a music piece that has been multi-played such as a CD. Has an aim.

[0005]

In order to achieve this object, an automatic transcription device of the present invention is a digital output device for emphasizing and outputting characteristics of each drum sound to be transcribed of an input music signal.
For each output of the filter group and the digital filter group, a power calculating means for calculating a time series of short-time power at constant time intervals, and a local portion of the time series of short-time power calculated by the power calculating means A local peak detecting means for detecting a peak is provided, and the time and the peak value of the local peak detected by the local peak detecting means are used as the hitting time and the strength of each drum.

[0006]

In the automatic music transcription device of the present invention having the above-mentioned structure, the digital filter group emphasizes the component of each drum sound in the inputted music signal. The power calculation means
A time series of short-time power is calculated for each output of the digital filter group at regular time intervals. The local peak detection means detects a time-series local peak of the power, and uses the time and peak value of the local peak as the striking time and strength of each drum.

[0007]

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 which constitutes the present embodiment inputs a played music signal and amplifies this signal to an appropriate voltage value.

The low-pass filter 2 suppresses aliasing distortion at the time of digitization 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 filter passing signal into a 16-bit digital signal with a sampling frequency of 12 kHz. I / O port 4
Connects the CPU 5, the A / D conversion device 3, and the display 8. The CPU 5 performs digital filter processing, power time-series calculation processing, peak detection processing, and the like on music signal data, and a RAM 6 and a ROM.
Connected to 7.

The RAM 6 stores the music signal data digitized by the A / D converter 3, the output of the digital filter, the time series of the power calculated by the CPU 5, the hit time of the drum sound detected by the CPU 5,
And an area for storing strength and the like. The ROM 7 stores digital filter processing logic, digital filter coefficients, power time series calculation logic, peak detection logic, and the like. The display 8 is for displaying processing results and the like.

FIG. 2 shows the automatic music transcription device of the present embodiment.
It is a functional block diagram with respect to the process of CPU5.

Each filter is composed of a finite interval impulse response (hereinafter, FIR) filter 21. Further, in order to have the characteristic of the corresponding drum sound, each filter uses the coefficient as a value obtained by normalizing the sample value in the section of a fixed time length from the rising time when the single sound of each drum sound is digitized. Figure 3 shows a 2400 point section (0.2
Second) minutes are displayed, and in the present embodiment, the values obtained by normalizing the sample values of 2400 points so as to satisfy the following equation are coefficients h ₀ , h ₁ , of the FIR filter 21a for bass drum transcription. ..., and is stored in the ROM 7 as h ₂₃₉₉ .

[0014] The FIR filter 21a is the A / D converter 3
FI obtained as described above for the music signal data x ₁ , x ₂ , ..., X _N (N is the number of samples of music signal data) digitized by
Filtering is performed using the R filter coefficient. Specifically, the outputs X ₁ , X ₂ , ..., X _N of the FIR filter
Can be calculated by the following equation.

[0015] The output of the thus obtained FIR filter responds significantly when the characteristics of the input music signal are similar to the characteristics of the bass drum sound used to determine the coefficient of the filter.

Similarly, FIG. 4 shows a snare drum output by a synthesizer (drum machine) and digitized, and is displayed for a 2400-point section (0.2 seconds) from the rising time. Sample values for these 2400 points are shown. Is used as a coefficient of the FIR filter 21b for snare drum transcription.

Further, the coefficients of other FIR filter 21 for drum transcription can be obtained by the same processing.

The power calculators 22a are respectively connected to the FIs.
For the outputs X ₁ , X ₂ , ..., X _{N of the} R filter 21, short-time power time series P ₁ , P ₂ , ... For each frame set at intervals of 120 points (10 msec).
., P _M is calculated. The short-time power P _m is calculated by the following equation.

[0019] The local peak detector 23a includes a smoother 24a and a differentiator 2
5a and a peak detector 26a.

[0020] The smoothing device 24a, when series P _1, P ₂ of the short-time power,..., With respect to P _M, time series Q ₁ of the smoothed power by the secondary-3-order polynomial fitting method, Q ₂ , ...
-Calculate Q _M. Calculation of the smoothed power Q _m by 2 and tertiary polynomial fit method using the 7 frames short-term power is a smoothing coefficient _{w -3 = -2, w -2 =} 3, w -1 =
6, w ₀ = 7, w ₁ = 6, w ₂ = 3, w ₃ = -2, the following equation is used.

[0021] The differentiator 25a uses the smoothing power time series Q ₁ ,
Q _2, ···, performs a differential treatment for the Q _M, time series △ Q ₁ of the time variation of the smoothed _{power, △ Q 2, ···, △} Q M
To calculate. In this embodiment, a linear regression coefficient based on the smoothing power for 7 frames as shown in the following equation is used as the time change amount.

[0022] The peak detector 26a detects a local peak in the time series of the smoothed power. For the local peak, a frame whose smoothing power and its time variation satisfy the following conditions is a peak frame.

ΔQ _l-1 > = 0, and ΔQ _{l + 1} <0 and Q _l > (certain threshold value) The peak detector 26a is the number of all peak frames satisfying the above conditions. 1 and the peak value Q _l are output and used as the striking time and intensity of the drum sound.

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

FIG. 5 is a flow chart showing the musical notation process for the bass drum performed in this embodiment.

First, the FIR filter 21a for bass drum transcription is composed of music signal data x ₁ , digitized by the A / D converter 3 and stored in the RAM ₆ .
x _2, · · ·, x against _N (N is the number of samples of data), the coefficient h ₀ of the FIR filter 21a stored in the ROM7 sought in the manner described above, h _1, · · ·, h
Filter processing is performed using ₂₃₉₉ (S1).

Here, the filtering process S1 will be described with reference to FIG. First, the CPU 5 initializes a data number n (hereinafter referred to as a data number) for processing music signal data (n = 1: S61). Next, RAM6
Using the music signal data stored in (1) as input, FIR filter processing is performed according to the following equation (S63).

[0028] Next, the CPU 5 increments the data number n and returns the process to S62 (S64).

In S62, the CPU 5 compares the data number n with the sample number N of the music signal data, and the judgment of n> N is "NO". The processes of S63 to S64 described above are repeated to make the judgment. If YES, process S
1 is ended (S62).

FIG. 9A shows an example of the music signal, in which the horizontal axis represents time and the vertical axis represents amplitude. The sample number N is 360,000 (30 seconds). (B) in FIG.
Is the output of the FIR filter 21a for bass drum notation with the horizontal axis representing time and the vertical axis representing amplitude as in (a), and it can be seen that the bass drum is emphasized.

Next, the power calculator 22a outputs the output X ₁ , of the FIR filter 21a for bass drum transcription.
120 points (10 ms for X ₂ , ..., X _N)
ec) The time series P ₁ , P ₂ , ..., P _M of short-time power is calculated for each frame set at intervals (S2).

Here, a short-time power time series calculation process S
2 will be described with reference to FIG. First, the CPU 5 initializes a frame number m (hereinafter, a frame number) for processing the output of the FIR filter 21a for bass drum transcription (m = 1: S71). Next, the outputs X ₁ , X ₂ , ..., X _N of the FIR filter for bass drum transcription stored in the RAM 6 are input, and the short-time power P _m is calculated by the following equation (S73).

[0033] Next, the CPU 5 increments the frame number m and returns the process to S72 (S74).

In S72, the CPU 5 compares the pointer 120 × m indicating the observation position with the sample number N of the output of the FIR filter 21a, and the determination of 120 × m> N is “N”.
In the case of "O", the processes of S73 to S74 described above are repeated. When the determination is "YES", the short time power calculation process S2 is ended (S72).

In FIG. 9, (c) shows the output from the power calculator 22a when (b) in FIG. 9 is input, with the horizontal axis representing time and the vertical axis representing the power value. The sample number M is 3000 (30 seconds).

Next, the smoother 24a calculates the smoothing power of the short-time power time series P ₁ , P ₂ , ..., P _M by the above-described second-order / third-order polynomial fitting method. Time series Q
₁ , Q ₂ , ..., Q _M are calculated. (D) in FIG. 9 is
Smoothing device 24a when (c) in FIG. 9 is input
The horizontal axis represents time, and the vertical axis represents power value. The sample number M is 3000 (30 seconds) (S3).

Next, the differentiator 25a is time-series Q _1, Q ₂ of the smoothed power, ..., performs differential processing for Q _M, the time series △ Q ₁ time change amount of the smoothed power , △
Q ₂ , ..., ΔQ _M are calculated by the above method. Figure 9
In (e), when (d) in FIG. 9 is input,
The output from the differentiator 25a is plotted with time on the horizontal axis and time variation on the vertical axis. The sample number M is 3000 (30 seconds) (S4).

Next, the peak detector 26a detects a time-series local peak of the smoothed power (S5).

Here, regarding the peak detection processing S5,
This will be described with reference to FIG. First, the CPU 5 sets the frame number m.
Are initialized (m = 1: S81).

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

ΔQ _m-1 > = 0, and ΔQ _{m + 1} <0 and Q _m > (certain threshold value) If the determination is “YES”, the CPU 5 executes the process S
If “NO” in S84, S84 is skipped and S8 is executed.
The process moves to 5 (S83).

In step S84, the CPU 5 writes the peak frame number m and the peak value Q _m in a predetermined address on the RAM 6 (S84).

Next, the CPU 5 increments the frame number m and returns the processing to S82 (S85).

In S82, the CPU 5 determines whether or not the processes of S83 to S85 have been completed for all the frames (m> M). If the determination is "YES", the entire process is completed.

(F) in FIG. 9 shows peaks detected by the peak detector 26a with respect to (e) in FIG.
A vertical line is drawn at the time corresponding to the frame.

The CPU 5 sets the time corresponding to the peak frame stored in the RAM 6 as the striking time of the bass drum sound and the peak value as its strength, and ends the bass drum notation processing.

The musical notation processing for other drum sounds can be performed by the same processing as the musical notation processing for the bass drum described above.

[0048]

As is apparent from the above description, the automatic transcription device of the present invention emphasizes the characteristics of each drum sound to be transcribed in the input music signal and outputs the digital filter group, and For each output of the digital filter group, a power calculating means for calculating a time series of short-time power at constant time intervals, and a local detecting peak of a time series of short-time power calculated by the power calculating means The drum sound can be automatically transcribed by providing a peak detection unit and using the time and the peak value of the local peak detected by the local peak detection unit as the striking time and strength of each drum.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an automatic music transcription device of the present invention.

FIG. 2 is a functional configuration diagram for a process of a CPU in the automatic music transcription device of the present invention.

FIG. 3 is an explanatory diagram showing sample values from a rising time of a bass drum sound.

FIG. 4 is an explanatory diagram showing sample values from a rising time of a snare drum sound.

FIG. 5 is a flowchart showing an operation of a musical notation process in the automatic musical transcription device of the present invention.

FIG. 6 is a flowchart showing an operation of a filter process in the automatic music transcription device of the present invention.

FIG. 7 is a flowchart showing an operation of a time series calculation process of short-time power.

FIG. 8 is a flowchart showing an operation of peak detection processing.

FIG. 9 is an explanatory diagram showing an output example of each process in the automatic music transcription device of the present invention.

[Explanation of symbols]

 21 FIR filter 22 Power calculator 23 Local peak detector

Claims (1)

[Claims] 1. An automatic transcription device for transcribing a drum sound from a multiple-played music, emphasizing the characteristics of each drum sound to be transcribed of an input music signal, and outputting the digital filter group and the digital filter group. .A power calculation unit that calculates a time series of short-time power at fixed time intervals for each output of the filter group, and a local peak that detects a local peak of the time series of short-time power calculated by the power calculation unit An automatic music transcription device comprising: a detecting unit, wherein the time and the peak value of the local peak detected by the local peak detecting unit are used as the striking time and strength of each drum.