JPH04181297A - voice recognition device - Google Patents

Abstract

PURPOSE:To accurately detect a speech section by providing a recognition part which recognizes the output of a speech section segmentation part as a speech and outputs the recognition result. CONSTITUTION:The periphery of the mouse of a speaker is photographed by a camera 1 and its signal is sent to a data conversion part 2; and the image of the periphery of the mouse is sampled to generate digital data, which is sent to and stored in an image storage part 3 and the difference from last image data is measured and sent to a speech section detection part 4. The output of the storage part 3 is monitored here and the signal is sent to the speech section segmentation part 8 until the state wherein the difference from the last data is less than a certain extent continues for a certain time. A pressure sensor 5 as an air flow detecting means detects the breathing pressure from the nose when the speaker voices words and sends it to an air flow detection part 6, which compares the output 5 of the pressure sensor with a reference value and sends the signal to the segmentation part 8 while the reference value is larger; and the the outputs of the detection parts 4 and 6 are received here and the speech from a microphone 7 is classified by sections and passed to the recognition part 9, which outputs the recognition result.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a speech recognition device. In a speech recognition device that detects and performs speech recognition.

As shown in Figure 2, by detecting a relatively low-level part of the signal input from the microphone, or by capturing the surrounding noise immediately before two utterances and comparing it with the signal level, the speech interval can be detected. was detected.

[Problems and objectives to be solved by the invention] However, conventional methods are easily affected by surrounding noise and cannot be used in places with high noise levels.
This system has been developed in view of these problems, and its purpose is to easily detect voice sections regardless of the level of surrounding noise, and to be able to handle a wide variety of utterances. The goal is to expand the range of applicability of speech recognition devices.

[Means for Solving the Problems] The speech recognition device of the present invention inputs the voices uttered by two speakers from a microphone, and performs speech recognition by detecting a speech section from the voice signal. an imaging means for photographing a moving image of the vicinity of the lips of the person, and digitizing the output of the imaging means to output digital data. A data conversion means stores the output of the data conversion means, compares it with the immediately previous data, and outputs the result.

The outputs of the image storage section and the storage section are monitored, and if the difference remains below a certain level as a result of the comparison for a certain period of time, a signal is output. This occurs when the voice section detection unit and the speaker 9 speak. It detects the airflow of exhaled air from the nose and outputs a signal. The airflow detection means and the output of the airflow detection means are compared with a reference value, and a comparison result is output. receiving outputs from the airflow detection section, the voice section detection section, and the airflow detection section;

to the audio data input from the microphone.

Data indicating the voice section is added and output. A speech section cutting section and the output of the speech section cutting section are recognized as speech, and a recognition result is output.

It is characterized by having a recognition section.

[Operation] To explain the operation of the speech recognition device of the present invention, the first
As shown in the functional block diagram in the figure, a camera serving as an imaging means (1) photographs the vicinity of the lips of a speaker (2), and the signal is sent to a data conversion unit (2). here.

The image near the speaker's lips is sampled into digital data and sent to the image storage section 3, where the digitized image is simultaneously stored.

The difference from the immediately previous image data is measured and the result is sent to the voice section detection section 4. In the voice section detection section 4.

The output of 3 is monitored, and the signal is sent to the audio section cutout section 6 until the difference with the previous data remains below a certain level for a certain period of time. On the other hand, the pressure sensor 5, which is the airflow detection means, detects the pressure of exhalation from the nose when the speaker 9 speaks, and sends it to the airflow detection section 6. The airflow detection section 6 compares the output of the pressure sensor 5 with a reference value, and sends a signal to the voice section extraction section 8 while the output is greater than the reference value.

The voice section extraction section 8 includes the voice section detection section 4 and the voice section detection section 6.
In response to the output of the airflow detection section, the audio input from the 7 microphones is divided into two sections.

The recognition unit 9 passes the output to the recognition unit 9, which recognizes the output of the speech segment extraction unit 8 and outputs a recognition result.

[Embodiment 1] The details of the speech recognition device of the present invention will be described below based on the illustrated embodiment.

FIG. 3 is a configuration diagram showing nine embodiments of the present invention.

Reference numeral 31 in the figure is a camera, and here a CCD camera is used. 32 is a data conversion means for digitizing the output of the camera 31. A/D converter 1
It is. The part surrounded by the broken line 33 is two image storage units, and the frame buffer control 331 [R:, the frame buffer 1 of 332, the frame buffer 2 of 333,
It consists of 334 and 335 exclusive OR (hereinafter referred to as XOR) gates and 336 OR gates.

34 is a voice section detection unit, which includes a counter 341;
It is composed of 342 comparators 1. 35 is a pressure sensor 6 The part surrounded by a dotted line 36 is an airflow detection section, and 361 is an amplifier and its output.

It is composed of 362 comparators 2 that compare the reference value determined by 363 variable resistors. 37
is a microphone. 38 is.

381 A/D converter 2 in the audio section cutting section.

382 buffer control circuit, 383 buffer 1,
It consists of 384 buffers 2 and 386 OR gates. 39 recognizes the voice and displays the result. This is the recognition part.

Next, the operation of the apparatus configured as described above will be explained based on the configuration diagram shown in FIG.

Camera 31 focuses on the lips of speaker 9.

Also, the camera is adjusted so that it does not photograph anything other than one mouth.

The video taken by the camera 31 is converted into digital data by the A/D converter 1 32.

The image data is sent to the image storage unit No. 33. This data is.

332 frame buffer 1 or 333 frame buffer 2. After the device starts operating, the image storage unit of 33 will be described.

Initial image data is stored in frame buffer 1 at 332. The next image data is stored in the frame buffer 2 of 333. At this time.

The frame buffer l of 332 is in the reading state, and the frame buffer l of 8
The marketed data is sent to 334 XOR gates. Reading frame buffer data is 32 A.
This is done in synchronization with the output of the /D converter 1. In addition, since the output of the A/D converter 1 of 32 is also input to the XOR gate of 334, the current image data and the immediately previous image data are XORed by the XOR gate of 334.
Then, it passes through the OR gate 336 and is sent to the voice section detection section 34. For subsequent image data, the roles of frame buffer 1 and frame buffer 2 are alternately switched between the current image and the above 6.

When XORed with the image just before that, the portion where there is no change in the two images, that is, the portion where there is no movement, is represented by O.

Parts with movement are represented by 1. Therefore, it is possible to determine whether or not there is movement in one image by counting the number of ■. Here, the counter 341 counts the number of 1's in the input data, and the counted result is sent to the comparator 1 342, where it is compared with a preset value (hereinafter referred to as a threshold level). As a result, a signal is output until the number l falls below the threshold level. In addition.

This threshold level is the camera or.

It depends on the noise level of A/D converter l. No. 35 pressure sensors are installed at positions that are sufficiently exposed to only the exhaled air from the nose of one speaker. The output of this pressure sensor is 3
It is input to the airflow detection section 6.

Thereafter, it is amplified by an amplifier 361, and compared with a reference value set by a variable resistor 363 by a comparator 2 362. As a result of the comparison, the output of the pressure sensor is
If it is larger than the reference value, a signal is output. 38 in the audio section extraction section.

The sound collected by microphone 37 is transferred to 381 and A.
/D converter 2 converts it into digital data.

383 buffers 1. Alternatively, it is stored in buffer 2 of 384. 382 buffer control circuit.

When receiving the signal from the voice section detecting section 34 or the signal from the airflow detecting section 36, the buffer 1. Alternatively, the audio data stored in the buffer 2 is output as one audio section data through the OR gate 386 to the recognition section 39. Normally, the signal from the audio section detection section 34 and the signal from the audio section detection section 36 are output.
The signals from the airflow detecting section are input to the voice section cutting section 38 almost at the same time.
When a voice consisting only of nasal sounds is produced manually, the lips of the speaker hardly move.Therefore, there is no signal from the voice section detection section 34, but the airflow detection section 36 detects the result of the nasal sound being uttered. It can detect airflow and output a signal. Therefore, even if the two uttered voices are only nasal sounds, the voice section can be detected. Immediately after the two devices start operating, the signal from the voice section detecting section 34 or the airflow detecting section 36 serves as a signal to start voice input. ,or,
When the first signal of 362 comparator 2 is input,
The sound from the time when the signal is manually input becomes the first sound and is stored in the buffer l.The buffer control circuit 382 controls the comparator 1.342. Alternatively, when the next signal from the comparator 2 of 362 is received, the voice data stored in the buffer l is read out and sent to the recognition section of 39.

At the same time, set buffer 2 of 384 to write state.
Next, the comparator l of 342, or.

3 until the signal is input from comparator 2 of 362.
Audio data is written to buffer 2 of 84.

These operations are switched every time there is an output from the comparator 1 of 342.

2. In the present invention, it is necessary for the camera to photograph only the movement of the lips. 6 Therefore, in the present embodiment, the fourth
As shown in the figure, 44 small CCD cameras are attached to arm 42, to which 43 sound collection units of the microphone device are attached, but they can also be integrated with the sound collection unit, or It can be attached anywhere as long as it moves in unison with the head.

In addition, in order to detect only the exhaled air from the nose of the second speaker, the pressure sensor 45 is further extended from the arm 42 in Fig. 4 and attached directly below the nose of the second speaker. It can be separated from the arm, or it can be attached to any part that can detect only the exhalation from the speaker's nose.

[Effects of the Invention 1 As explained above, 2 According to the present invention, 2 accurate noise can be achieved without being affected by the surrounding noise level at all.

It is possible to detect voice sections and further expand the range of voices that can be handled by the voice recognition device, and the effect is remarkable.

[Brief explanation of drawings]

FIG. 1 is a functional block diagram showing the configuration of a speech recognition device according to the present invention. FIG. 2 is a functional block diagram showing a conventional example, and FIG. 3 is a configuration diagram showing two embodiments of the present invention. FIG. 4 is a conceptual diagram showing an example of how to attach a camera, a pressure sensor, and a microphone according to the present invention. 1...Imaging means 2...Data conversion section 3...Image storage section 4...Audio section detection section 5...Airflow detection means 6...Airflow detection section 7...Microphone 8...・Voice section cutting section 9...Recognition section 21...Microphone 22...Speech level detection section 23...Voice section cutting section 24...Recognition section 31...Camera 32...A/D Converter 1 33... Image storage unit 331... Frame buffer control circuit 332... Frame buffer 1 333... Frame buffer 2 334, 335...-XOR gates 336, 366... OR gate 34...・Voice section detection unit 341...Counter 342...Comparator 1 35...Pressure sensor 36...Air flow detection unit 361...Amplifier 362...Comparator 2 363...Variable resistor 37...・Microphone 38...Voice section cutting unit 381...A/D converter 2 382...Buffer control circuit 383...Buffer 1 384...Buffer 2 39...Recognition unit 41...Speech Person's mouth 42...Arm 43 of the microphone device...Sound collection unit 44...CCD camera 45...Pressure sensor or more Applicant: Seiko Epson Co., Ltd. Agent Patent attorney Kizobe Suzuki and 1 other person Figure 1

Claims (1)

[Scope of Claims] a) A voice recognition device that performs voice recognition by inputting the voice uttered by a speaker from a microphone and detecting a voice section from the voice signal, b) A moving image of the vicinity of the speaker's lips. c) data converting means for digitizing the output of the image capturing means and outputting digital data; d) storing the output of the data converting means and comparing it with the immediately preceding data. e) an image storage unit that monitors the output of the storage unit and outputs a signal if the difference remains below a certain level as a result of the comparison for a certain period of time; f) airflow detection means for detecting the airflow of exhaled air from the nose that occurs when the speaker speaks and outputting a signal; g) comparing the output of the airflow detection means with a reference value; an airflow detection section that outputs a comparison result;
A speech segment cutting unit that adds and outputs data indicating a speech segment; and i) a recognition unit that recognizes the output of the speech segment cutting unit as speech and outputs a recognition result. Speech recognition device.