JPS6356698A - Numerous speaker voice recognition equipment - Google Patents

Abstract

(57) [Abstract] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a multi-speaker speech recognition device that recognizes the voices of a large number of speakers.

(Prior Art) FIG. 3 is a block diagram showing a conventional multi-speaker speech recognition device. In the figure, 31 is a one-wooden microphone that picks up voices emitted by many speakers and outputs audio signals;
2 is an amplifier (hereinafter abbreviated as AMP) for amplifying the audio signal from the microphone 31; 33 is provided for each speaker, and when pressed only by the speaker who made the sound, a standard pattern selection unit 34 (described later) is selected. A discrimination switch 34 selects a standard pattern pre-registered in the standard pattern section 35, which will be described later, based on a signal from the discrimination switch 33. 35 is a standard pattern section in which standard patterns for each speaker are registered in advance, and 36 is a standard pattern section for outputting the standard pattern from the audio signal from the AMP 32 and the selected standard pattern section 35. This is a recognition unit that performs identification with standard patterns.

Next, the operation of the conventional multi-speaker speech recognition device will be explained using FIG.

First, since the speaker who uttered the voice and the standard pattern of that speaker are registered in the standard pattern section 35 in correspondence, when a standard pattern is created from the uttered audio signal and registered in the standard pattern section 35, the speech The user presses the discrimination switch 33 to emit a sound.

Then, for example, when the speaker ■ speaks, the speaker ■ presses the discrimination switch 33 of the speaker ■. Then, the standard pattern selection section 34 selects the standard pattern A of the speaker ■ from the standard pattern section 35 based on the signal from the discrimination switch 33, and supplies it to the recognition section 36. Therefore, the speech uttered by the speaker ■ is compared with the standard pattern of the speaker ■ from the standard pattern unit 35 in the recognition section 36, identification is performed, and the speech is recognized. Incidentally, when the speaker (2) continues to speak, there is no need to press the discrimination switch 33.

(Problem to be Solved by the Invention) However, in the conventional device described above, the speaker who has uttered the voice is detected by pressing the respective discrimination switch, and the standard pattern corresponding to that speaker is selected and recognized. In this case, each speaker must have his or her discrimination switch in hand,
In addition, there were some inconveniences in operability, such as having to press a discrimination switch every time a voice was emitted.

The present invention is intended to solve these problems, and it is an object of the present invention to provide a multi-speaker speech recognition device with excellent operability.

(Means for Solving the Problems) In order to solve the above-mentioned problems, the present invention provides the following means in a multi-speaker speech recognition device that recognizes the voices of a plurality of speakers.

A first invention includes a first means for installing a plurality of microphones that convert a speaker's voice into an audio signal at a distance, and identifying the speaker's position based on each audio signal from each microphone; i2 means for detecting a speaker based on the identification result from the first means and selecting a pre-registered standard pattern corresponding to the detected speaker; and a standard selected by the second means. and third means for recognizing the speaker's voice based on the pattern.

A second invention includes a first means for installing a plurality of microphones spaced apart from each other to convert the voice of the speaker into an audio signal, and identifying the position of the speaker based on each audio signal from each microphone; a second means for extracting the frequency characteristics of the speaker's voice and identifying them with frequency characteristics previously established for each speaker; and based on each identification result from the first means and the second means. a third means for detecting a speaker and selecting a pre-registered standard pattern corresponding to the detected speaker;
and fourth means for recognizing the speaker's voice based on the standard pattern selected by the third means.

(Operation) According to the first and second inventions having the above-described configurations, each operates as follows.

In the first invention, the speaker's position is first identified based on each audio signal from each of the plurality of microphones installed at a distance. Then, the speaker is detected based on this identification result, and a pre-registered standard pattern corresponding to the detected speaker is selected. Therefore, the speaker's voice is recognized based on this selected standard pattern.

The second invention differs from the invention of 7gl in that when detecting a speaker, frequency characteristic identification of the speaker's voice is added in addition to the speaker position identification as in the first invention. This makes it possible to reliably identify the -layer speaker.

Therefore, the present invention can solve the above problems and provide a multi-speaker speech recognition device with excellent operability.

(Example) Hereinafter, one embodiment of the present invention will be described based on the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention. In the figure, 1 and 2 are microphones that pick up voices emitted by many speakers and output audio signals, 3 is an AMP that amplifies each voice signal from microphones 1 and 2, and 4 is an AM
A recognition unit identifies the audio signal from P3 and a standard pattern from a standard pattern unit 8, which will be described later. 5 is a pitch extraction unit that calculates the average pitch frequency of the audio signal from AMP 3. 6 is a recognition unit for each of the signals from the microphone 1.2. a position identification unit 7 that calculates the power difference between the voice signals and registers the difference as position information of each speaker, and identifies the position data of the speaker who actually uttered the voice and the registered position information; Pitch extractor 5
A standard pattern corresponding to the speaker is selected from among the standard patterns registered in advance in the standard pattern section 8, which will be described later, based on the speaker identification output based on the average pitch frequency from the speaker and the position identification output from the position identification section 6. 8 is a standard pattern section for pre-registering standard patterns for each speaker; 9 is a key for manually inputting the standard pattern registration name for each speaker; 10 is key 9
The standard pattern registration name manually created from the standard pattern section 8
This is a recognition control unit that registers the standard pattern along with the standard pattern to be registered.

Next, the operation of this embodiment will be explained using FIG.

First, create a correspondence list between speakers and standard patterns. Using the key 9, the standard pattern registered name of each speaker (A, B, C, etc.) is entered manually in the order from microphone 1 to microphone 2, for example. Next, microphone 1,
Turn on 2 and switch from microphone 1 to microphone 2.
In this order, the outside speaker utters several words at intervals. The position identification unit 6 includes the microphone 1. The position of each speaker is determined based on the power difference between the audio signals input manually from the microphone 2, and is registered as position information of each speaker.

At this time, if there are a large number of speakers or if the speakers are close to each other, there is a possibility of incorrect position determination, so the average pitch frequency of each speaker is determined and registered by the pitch extraction section. This average pitch frequency is the vibration frequency of the vocal cord sound source.

In the case of Japanese dogs, adult males have an average frequency of 100 to 150 Hz, females have an average frequency of 250 to 300 Hz, and children are somewhere in between, so it is relatively easy to separate the sexes. By determining the average pitch frequency of each speaker, erroneous identification of nearby speakers can be further prevented than by position identification alone. If the average pitch frequencies are also similar, the speaker position can be changed. With the above steps, speaker location registration is completed.

During the recognition operation, audio signals input from the microphones 1 and 2 are input to the pitch extraction section 5 and the position identification section 6, and the position identification output based on the power and the speaker identification output based on the average pitch frequency are sent to the standard pattern selection section 7. Manpower. As shown in FIG. 2, the standard pattern selection section 7 performs position identification output based on power and speaker identification based on average pitch frequency output, and selects a standard pattern corresponding to the speaker. In FIG. 2, it is assumed that the speaker is located between the microphones 1.2 in the position distribution as shown in the figure.

At this time, the audio signals from microphones 1 and 2 are
The pitch extraction unit 5 and position identification unit 6 in the figure
Assuming that the position distribution as shown in the figure is calculated, the overlapping part of the two position distributions is determined to be the speaker's position, and the speaker (speaker ■ in the figure) is identified.

Then, the recognition unit 4 performs a recognition operation by identifying the selected standard pattern and the audio signal from the microphone 1.2.

(Effects of the Invention) As explained above, according to the present invention, the speaker can be automatically and accurately detected based on the position identification of the speaker using human-generated audio signals from a plurality of microphones, and the detected speaker can be heard. It is possible to perform speech recognition by selecting a standard pattern corresponding to the speaker, and to identify the speaker even more accurately by extracting the frequency characteristics from the speech signal emitted by the speaker. , it is possible to provide a multi-speaker speech recognition device with excellent operability.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing an example of position distribution according to this embodiment, and FIG. 3 is a block diagram showing a conventional multi-speaker speech recognition device. 1.2...Microphone, 3...Amplifier, 4...
- Recognition section, 5... Pitch extraction section, 6... Position identification section, 7... Standard pattern selection section, 8... Standard pattern section, 9... Key, 10... Recognition control section .

Claims (2)

[Claims] (1) In a multi-speaker speech recognition device that recognizes the voices of multiple speakers, multiple microphones that convert the voices of speakers into audio signals are installed at a distance, and the a first means for identifying the position of the speaker; and detecting the speaker based on the identification result from the first means, and selecting a pre-registered standard pattern corresponding to the detected speaker. A multi-speaker speech recognition device comprising a second means and a third means for recognizing the speech of speakers based on the standard pattern selected by the second means. (2) In a multi-speaker speech recognition device that recognizes the voices of multiple speakers, multiple microphones that convert the voices of the speakers into audio signals are installed at a distance, and the a first means for identifying the position of the speaker; a second means for extracting the frequency characteristic of the speaker's voice and identifying it with a frequency characteristic registered in advance for each speaker; and the first means. and a third means for detecting a speaker based on each identification result from the second means and selecting a pre-registered standard pattern corresponding to the detected speaker; A multi-speaker speech recognition device comprising: fourth means for recognizing the speech of the speakers based on the standard pattern selected by the method.