JPH0736483A - Method and device for sound separation - Google Patents

Abstract

PURPOSE:To provide the sound separating method and device which individually and adaptively handle sounds, one by one. CONSTITUTION:The sound separating device is equipped with a sound input terminal 101 for inputting a sound signal, sound signal extraction parts 102-104 which extract only specific sound signals, sound signal tracking parts 106 and 107 which track the specific sound signals, and a sound signal quantity control part 105 which generates the sound signal extraction parts and sound signal tracking parts. By this sound separating method, a sound signal extraction part and a sound signal tracking part corresponding to a new sound are generated when the sound signal quantity control part 105 detects the new sound being generated in the inputted sound signal and a sound signal extraction part and a sound signal tracking part corresponding to a sound are eliminated when the sound disappears from a mixed sound.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sound separation method and device, and more particularly to a sound separation method and device for adaptively extracting and separating the sound contained therein from a mixed sound in which a plurality of sounds are arbitrarily generated and extinguished. Regarding

[0002]

2. Description of the Related Art A conventional example of a sound separating apparatus will be described with reference to a sound separating apparatus shown in FIG. Feature extraction unit 2
02, 203, 204 and 205 are for converting the acoustic input 201 into a time series of feature vectors based on different features. The voice separation unit 206 separates voices by grouping the time series of feature vectors input from each feature extraction unit based on the feature grouping rules of voices.

[0003]

In the conventional example of the voice separation device described above, the voice is separated when all the preset feature extraction processing is completed, so that, for example, an emergency is notified during the mixed sound. It was unsuitable as a voice separation device to be used in an actual environment where a warning sound is included.

The present invention provides a sound separating method and apparatus which solve the above problems.

[0005]

SOLUTION: A sound input terminal 101 for inputting a sound signal is provided, and sound signal extraction units 102, 103, 104 for extracting only a particular sound signal are provided, and sound for tracking a particular signal is provided. An acoustic separation device including the signal tracking units 106 and 107 and the acoustic signal extraction unit and the acoustic signal number control unit 105 that generates the acoustic signal tracking unit is configured. Then, when the sound signal number control unit 105 detects that a new sound is generated in the input sound, the sound signal extraction unit and the sound signal tracking unit corresponding to this sound are generated,
The sound signal extraction unit and the sound signal tracking unit corresponding to the sound when the sound disappears from the mixed sound also constitute a sound separation method that causes the sound to disappear.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described with reference to FIGS. 1, 2 and 3. Here, a case will be described where a sound is separated from a mixed sound in which a plurality of sounds having a harmonic overtone structure exist. A sound having a harmonic overtone structure means a sound in which one sound is composed of a sound having a fundamental frequency component and a sound having a component having a frequency that is an integral multiple of the fundamental frequency component.

First, referring to FIG. 2, FIG. 2 shows the sound separating apparatus of the present invention constructed by a high speed electronic computer in a silent state. Assuming that one sound is newly input to the sound input terminal 301 from the silent state, the sound signal extraction unit 302 outputs the input as it is to the sound signal number control unit 305. The acoustic signal number control unit 305 obtains the intensities E of several harmonic groups having the fixed frequency ω as the fundamental frequency in the input sound based on the following formula.

[0008]

[Equation 1] Here, t: time i: i-th overtone component x (t): time t
Acoustic input at ω: fundamental frequency.

If the overtone of the maximum intensity among the obtained overtone intensities exceeds a certain threshold, the acoustic signal number control unit 3
Reference numeral 05 detects the presence of a sound, and generates one acoustic signal extraction unit 403 and one acoustic signal tracking unit 406 for tracing overtones having a fundamental frequency near the fixed frequency, as shown in FIG. The acoustic signal tracking unit 406 thereafter performs tracking by obtaining the fundamental frequency that maximizes the equation (1) for the acoustic input. The acoustic signal tracking unit 406 also transmits the component of the acoustic signal tracked by itself as a subtraction signal to the acoustic signal extraction unit 402. In the subtraction signal, first, with respect to the fundamental frequency ω that maximizes the expression (1), the strength A _i and the phase φ _i of each overtone component are obtained by the following expressions, and then the period and the strength A _{i of} each overtone component It is obtained by adding the sine wave formed from the phase φ _i and all overtones.

[0010]

[Equation 2] Since the acoustic signal extraction unit 402 subtracts the subtraction signal from the input and outputs the subtracted signal, the newly generated acoustic signal tracking unit 406.
The acoustic component corresponding to
Will not be entered in. A mixed sound of a plurality of sounds is input, and a plurality of acoustic signal extraction units 102, 103 and 10 are input.
4 and the acoustic signal tracking units 105, 106 and 107 are generated, the state of FIG. 1 is one acoustic signal tracking unit 106.
Transmits the subtraction signal to all the acoustic signal extraction units 102 and 104 other than the acoustic signal extraction unit 103 corresponding to itself.
The method of transmitting the subtraction signal is common to all other acoustic signal tracking units. Therefore, all the acoustic signal extraction units 10
2, 103 and 104 extract corresponding acoustic signals by subtracting the acoustic components other than the acoustic components to be extracted from themselves as the subtracted signals from the input.

When one sound is deleted from the mixed sound, the value of the equation (1) becomes smaller than the above-mentioned threshold value in the acoustic signal tracking unit 106 corresponding to this sound. Judge that it has failed. Here, the corresponding acoustic signal extraction unit 103 and acoustic signal tracking unit 106
Disappears. A case where the above-described sound separation method and apparatus are applied to separation of a mixed sound composed of sounds generated by a plurality of sound sources will be described with reference to FIGS. 4 and 5.

FIG. 4 is a diagram showing a temporal change 501 of the fundamental frequency of a male voice given as an input and a temporal change 502 of the fundamental frequency of an interfering sound in a mixed sound. The acoustic input is a mixed sound in which a male voice "Aiueo" and an interfering sound with a fundamental frequency of 200 Hz are superposed, and this is sampled at 12 kHz and AD converted with 16 bits to 30 ms.
A frame period of 7.5 is used.

It can be understood from FIG. 5 that the disturbing sound is removed and only the fundamental frequency of the male voice and its harmonic components are separated.

[0014]

As described above, the sound separation method and apparatus of the present invention respond to the sound when the sound signal number control unit 105 detects that a new sound is generated during the input sound. Generate an acoustic signal extraction unit and an acoustic signal tracking unit that do this, and extract this sound until it disappears from the mixed sound,
When the sound is tracked and disappears, the sound signal extraction unit and the sound signal tracking unit corresponding to this sound are made to disappear, and each sound can be dealt with individually and adaptively.

[Brief description of drawings]

FIG. 1 is a diagram for explaining an embodiment of a sound separating method and device according to the present invention.

FIG. 2 is a diagram illustrating an embodiment of a sound separation method and device according to the present invention.

FIG. 3 is a diagram for explaining an embodiment of a sound separation method and device according to the present invention.

FIG. 4 is a diagram showing a temporal change of a fundamental frequency of a male voice given as an input and a fundamental frequency of an interfering sound in a mixed sound.

FIG. 5 is a diagram showing a temporal change of a fundamental frequency of a male voice separated by removing an interfering sound.

FIG. 6 is a diagram illustrating a conventional example of an acoustic separation device.

[Explanation of symbols]

 101 Acoustic Input Terminals 102, 103, 104 Acoustic Signal Extraction Units 106, 107 Acoustic Signal Tracking Unit 105 Acoustic Signal Number Control Unit

Claims (2)

[Claims] 1. An acoustic signal input section for inputting an acoustic signal, an acoustic signal extraction section for extracting only a specific acoustic signal, and an acoustic signal tracking section for tracing a specific acoustic signal. An acoustic separation device comprising an acoustic signal number control unit for generating an extraction unit and an acoustic signal tracking unit. 2. When a sound signal number control unit detects that a sound is newly generated in an input sound, a sound signal extraction unit and a sound signal tracking unit corresponding to the sound are generated, and a certain sound is generated. A sound separation method characterized in that, when the sound is removed from the mixed sound, the sound signal extraction unit and the sound signal tracking unit corresponding to this sound are made to disappear.