JPH046600A - Voice recognition device - Google Patents

Abstract

PURPOSE:To increase the recognition rate by incorporating a voice input part, a feature extraction part, a recognition and decision part, and a recognition result transmission part in a wrist watch. CONSTITUTION:A voice input part consists of a microphone, a high-frequency- range emphasizing filter, and an AD converter and samples a voice at an 8kHz, 12-bit rate. The voice signal which is converted into a digital signal is converted by the feature extraction part into frequency elements and feature parameters in the frequency range of the voice are extracted as a voice spectrum which is the converted signal. Further, the feature parameters which are obtained are compared with by the recognition and decision part with the feature parameters of the phoneme of the speaker in the frequency range which are learnt in advance to recognize and decide which phoneme is voiced, thereby coding the voice. Then the voice input part, feature extraction part, recognition and decision part, and recognition result transmission part are incorporated in the wrist watch and a recognition result reception part and a recognition result processing part are incorporated in a system main body. Consequently, the recognition rate is increased.

Description

[Detailed description of the invention] [Industrial application field 1 The present invention relates to a speech recognition device using a wristwatch.

[Conventional technology 1 In the conventional technology, materials from the Batan study group of the Institute of Electronics and Communication Engineers, PR
A speech recognition device for an unspecified speaker, which is composed of a speech input section, a feature extraction section, a recognition determination section, and a recognition result processing section, is known, such as the conversation speech recognition system described in No. L79-61.

[Problems and Objectives to be Solved by the Invention 1 However, in the conventional technology, in a device that needs to recognize the voice of an unspecified speaker, in order to recognize the voice of an unspecified speaker, a feature extraction unit, The structure of the recognition determination unit is extremely complicated, the hardware for performing real-time processing is large-scale and expensive, and the recognition rate is low.

The recognition rate of currently existing speech recognition devices for non-specific speakers is about 95% at best, and has the problem that erroneous recognition is certain to occur.

This is due to the fact that human voice waveforms vary greatly from person to person.

Therefore, the present invention is intended to solve these problems, and its purpose is to use each individual's wristwatch in a device that needs to recognize the voice of an unspecified speaker.
By recognizing the voice of a specific speaker, the recognition rate can be increased and the size of the voice recognition unit can be reduced.

[Means for solving the problem 1: In a speech recognition device that inputs the voice of a specific speaker, operates a machine based on the input voice, and performs the processing desired by the speaker, the voice is manually input and the voice is a voice input unit that converts the signal into a digital signal; a feature extraction unit that receives the signal from the voice input unit and extracts its characteristic parameters; and a feature extraction unit that receives the signal from the feature extraction unit and converts the signal into a voice code such as a phoneme. a speech recognition unit for recognition; a recognition result transmission unit that receives a signal from the speech recognition unit and sends the signal;
a recognition result receiving section that receives a signal from the recognition result transmitting section; a recognition result processing section that receives the signal from the recognition result receiving section, analyzes the meaning, and takes a predetermined response; The present invention is characterized in that the feature extraction section, the recognition determination section, and the recognition result transmission section are built into a wristwatch.

[Example] An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a system configuration diagram of a speech recognition device according to the present invention.

As shown in FIG. 1, the speech recognition device of the present invention includes a speech input section, a feature extraction section, a recognition determination section, a recognition result transmission section,
Consisting of a recognition result receiving section and a recognition result processing section, the voice input section, feature extraction section, recognition determination section, and recognition result transmitting section are built into the wristwatch, and the recognition result receiving section and recognition result processing section are integrated into the system. It has a structure built into the main body. Here, the voice recognition performed inside the wristwatch is a specific speaker recognition in which only the voice of the individual owner of the wristwatch is recognized.

Each part will be explained below. The audio input section consists of a microphone, a high-frequency emphasis filter, and an AD converter, and outputs audio at 8KH.
z, sample at 12 bits. The audio signal converted into a digital signal is converted into a frequency dimension in a feature extraction section, and feature parameters in the frequency domain of the audio are extracted from the audio spectrum that is the converted signal. The obtained feature parameters are processed in the recognition determination unit.
By comparing the frequency-domain feature parameters of the phonemes of the speaker (i.e., the owner of the watch) that have been learned in advance, it recognizes and determines what kind of phoneme was uttered as speech, and performs speech encoding. .

FIG. 3(a) shows the original waveform 1 of the vowel /a/ uttered by a certain man. Here, 2 is the time axis, and one scale section is 10 m5. FIG. 3(b) shows an audio spectrum envelope 3 obtained by converting the frequency of the audio signal. Here, the horizontal axis 4 is a frequency axis, and one scale section is IKHz. Moreover, the vertical axis 5 is the amplitude (dB). The characteristics of the voice spectrum envelope 3 shown here have extremely high reproducibility for each individual. Therefore, speech recognition of a specific speaker is
The frequency characteristics of each phoneme of the individual are stored in the memory of the speech recognition unit in advance as learning button data, and during recognition, the learning button data that matches the frequency characteristics of the uttered voice signal is selected. be able to recognize phonemes. Therefore, since speech recognition in the present invention can be achieved with a very simple algorithm,
The hardware configuration for performing processing in real time is also extremely simple, making it possible to downsize the entire device. Furthermore, in specific speaker recognition, it is possible to obtain a recognition rate close to 100% even with a simple device such as the present invention.

The voice code strings obtained through real-time processing are transmitted via radio waves from the recognition result transmitter inside the wristwatch at any time, and are received by the recognition result receiver in the system main body. The received voice code string is sent to the recognition result processing section, where various processes necessary for each system are performed.

As an example, an example in which the speech recognition device of the present invention is applied to a ticket vending machine will be described. The speaker looks at his watch and records the name of his destination station, train time, number of tickets, and whether or not he has a reserved seat.
etc., the system receives the information, reserves seats on the train, calculates the price, and sells tickets quickly and accurately.

By incorporating the voice recognition unit inside the wristwatch, the speaker can specify the desired ticket while checking the current time on his or her wristwatch, making it possible to purchase tickets more quickly and accurately. .

Furthermore, by adding a structure to this wristwatch to memorize necessary train times, etc., it is possible to further improve the efficiency of ticket purchasing.

This voice recognition device can also be applied to vending machines for juice, cigarettes, etc. Furthermore, by commonizing and processing the signals transmitted from the wristwatches, it becomes possible to operate all the devices using the wristwatches owned by each person. In other words, it becomes possible to purchase any product anywhere, by voice, using a wristwatch.

Furthermore, by adding a structure that connects the signal from the recognition result transmitting section of the wristwatch as sound waves or light to a telephone line, it is possible to send and receive extremely large amounts of information by voice using a telephone line, making it possible to do so easily. This can be done using a voice recognition device.

[Effects of the Invention] As explained above, the voice recognition device of the present invention has a structure in which the voice input section, feature extraction section, recognition determination section, and recognition result transmission section are incorporated into the wristwatch, so that it can be used for unspecified speech. In a device that needs to recognize a person's voice, processing can be performed by specific speaker voice recognition, which has the effect of greatly increasing the recognition rate.

In particular, when the speech recognition device of the present invention is applied to processing that is performed while checking the current time, it has the effect of enabling the processing to be performed quickly and accurately.

Furthermore, since the signal sent from the wristwatch is a voice code, it has the effect of compressing the signal, thereby simplifying the communication method and downsizing the device.

[Brief explanation of the drawing]

FIG. 1 is a system configuration diagram of a speech recognition device of the present invention. FIG. 2 is a system configuration diagram of a conventional speech recognition device. FIG. 3(a) is an original waveform diagram of the vowel /a/ uttered by a certain man, and FIG. 3(b) is a speech spectrum envelope diagram obtained by converting the frequency of the audio signal of the vowel /a/. Applicant Seiko Epson Co., Ltd. Agent Patent Attorney Kizobe Suzuki (and 1 other person) Kunijiji G1 Figure 1 Figure 2

Claims (1)

[Scope of Claims] (a) A speech recognition device that inputs the voice of a specific speaker, operates a machine based on the input voice, and performs processing intended by the speaker, (b) , an audio input section that converts the audio signal into a digital signal, (c) a feature extraction section that receives the signal from the audio input section and extracts its feature parameters, (d) receives a signal from the feature extraction section, a speech recognition section that recognizes the signal as a speech code such as a phoneme; (e) a recognition result transmission section that receives the signal from the speech recognition section and sends the signal; (f) a recognition result transmission section that receives the signal from the recognition result transmission section. (g) a recognition result processing unit that receives the signal from the recognition result reception unit, analyzes the meaning, and takes a predetermined response; (h) the voice input unit; the feature extraction unit; , the recognition determination section, and the recognition result transmission section are incorporated into a wristwatch.