JPS59147398A - Voice section detector - Google Patents

Abstract

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

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech interval detection device suitable for use in an audio signal processing device such as speech recognition.

Conventional techniques and 11. Audio signal processing devices for the purpose of speech recognition, etc. are already well known, but in this audio signal processing device, it is necessary to detect a time interval in which speech exists as a preprocessing. In order to do this, it is important to clearly understand when the audio starts and ends.

For this preprocessing, conventionally, for example, an amplitude level comparator or the like is used to recognize the point in time when the input signal from the microphone exceeds a certain threshold value as the start point, and the point in time when it falls below a certain threshold value as the end point. was.

However, since the voice signal is generally weighted by ambient noise, the start and end detection times vary depending on the noise level, making it difficult to detect the correct voice section.

1-Once The present invention has been made to solve the above-mentioned drawbacks of the prior art, and in particular, by combining lip image processing in addition to processing the time waveform of audio input, It is possible to correctly detect the voice section of .

The configuration of the present invention will be described below based on embodiments.

FIG. 1 is a configuration diagram for explaining one embodiment of the present invention, in which 1 is a speaker and 2 is a microphone.

3 is an image sensor, and as shown in the figure, a type of image sensor 3 is placed on the microphone 2 that collects the audio wave A uttered by the speaker 1, and the image sensor 3 is used to collect the audio signal. At the same time, lip image information of the speaker is also collected.In principle, any type of image sensor 3 may be used as the image sensor 3 used here, but considering real-time processing performance, electric It is desirable to have something that can be converted into a signal, and not to use photographic plates that use emulsions or the like.Therefore, an image pickup plate that uses a small vidicon or CCD sensor may be considered.
In the present invention, since high resolution is not required, a solid-state imaging device such as a CCD board having a relatively small number of elements is suitable. The present invention uses the above-described image sensor to obtain an image of the lips of a speaker, and thereby detects a speech interval. In other words, in the utterance period, the speaker's [1] lip is constantly moving, and the lip image information is also constantly changing, but in the non-utterance period, the speaker's lips are constantly changing.
The 11th lip image also remains constant. Therefore, - Generally speaking, in the unvoiced section, 11 images show a high frame correlation value, while in the vocal section, 45
In this case, [1 lip image has a relatively low frame correlation value, and paying attention to this fact, the frame correlation value is +11
After maintaining a high value for 11.7 seconds, it sharply decreases from high to low] 1.7 points can be determined as the starting point of the voice, and on the other hand,
Frey, the correlation value increases dramatically from low to high, and then for a long time,
If it seems to maintain a high value, the point of change can be separated from the point of voice start. In this case, in order to accurately detect the voice section, the period (l rn s
p c: ~ several Ill S e (: degree) h; essential; It is clear that it can be done. In addition, monthly norm correlation 1
However, it cannot be said that 4L is necessarily the best for practical purposes. 3L
It is also possible to do so.

FIG. 2 is a block diagram of main parts showing another embodiment of the present invention. In the figure, 4 is an objective lens, 5 is an optical fiber, 6 is an optical sensor, and 7 is a scanner. Therefore, in this embodiment,
In order to make the portion integrated with the audio wave pickup microphone in the lip image detection section small and lightweight,
The objective lens 4 is separated from the image sensor unit 6, and only the objective lens 4 is integrated into a microphone, and the image sensor unit 6 is placed in front of the main processing unit, and an optical fiber 5C is connected between them. This is what I did.

It should be noted that each of the embodiments of the present invention (A and β2) has been described above, but the present invention is not limited to two embodiments. Difference No. 1; Depends (
It is also considered to provide a means for correcting the Hrl deviation when it cannot be irradiated.

According to the present invention, it is clear from the explanation of Yo-ichi-Kanei.
Since the audio signal and the lip image signal are used as information sources for audio processing, it is possible to detect the audio segment (■) with good viscosity.

[Brief explanation of the drawing]

FIGS. 1 and 2 are configuration diagrams for explaining an implementation 911 of the present invention, respectively. 1...Speaker, 2...Microphone, 3...Picture (f
if (sensor, 4... objective lens, 5... optical fiber <-6... image sensor, 7... scanner. Figure l

Claims (1)

[Claims] , has an image sensor that obtains an image of the speaker's lips by integrating it with a mask logo that pinks up the audio waves in box 15, and uses the speaker's audio waves and lip image information as an information source for audio processing such as speech recognition. A voice section detection device characterized by