JPH04181998A - Device and method for speech recognition - Google Patents

Abstract

PURPOSE:To recognize the break of the head or tail of a voiced word and the mixture of a circumferential noise into the voiced word by comparing numeric data on an input speech signal with stored numeric data on standard patterns and recognizing the speech. CONSTITUTION:A speech input part 11 amplifies the input speech signal and generate numeric data by an A/D converter, etc., and a speech output part 12 converts the standard patterns stored in a storage part 14 into a speech signal by a D/A converter, etc., and outputs it. Then a control recognition part 13 controls respective functions and compares the numeric data inputted by the speech input part 11 with the standard patterns stored in the storage part 14 to perform a recognizing process. Further, when a standard pattern is registered, the control recognition part 13 stores the voice signal made into the numeric data by the voice input part 11 in the storage part 14 as the standard pattern with the command of an operation command part 15 according to a program in the storage part 14 and the voice output part 12 makes the standard pattern into a speech signal. Consequently, the break of the head or tail of the registered standard pattern, the mixture of the circumferential noise, etc., can be confirmed auditorily.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a speech recognition device and method, and particularly to a speaker-specific speech recognition device and method.

[Conventional technology]

Registration of a standard pattern in a conventional speaker-specific speech recognition device is performed by converting the standard pattern into numerical data by the control command unit 22 or the voice input unit 21 according to the commands from the operation command unit 24 and the program in the storage unit 23 shown in FIG. The recorded audio signal is stored in the storage unit 23.
It is only stored in .

[Problem to be solved by the invention]

With the standard pattern registration method of the conventional speech recognition device described above, it is not known how the input voice was stored in the device, so it is difficult to judge whether the standard pattern is good or bad, which directly affects the recognition rate. The problem is that you have no choice but to make a judgment and use your own experience.

In other words, it has the disadvantage that it is not possible to confirm whether the first four or last words of a spoken word are cut off, or whether surrounding noise is mixed in with the spoken word.

[Means to solve the problem]

The speech recognition device of the present invention includes (a) means for converting an input speech signal into numerical data, (b) means for storing this numerical data as a standard pattern, and (c) numerical values of the stored standard pattern. (d) A means for comparing and recognizing the numerical data of the input audio signal with the numerical data of the stored standard pattern; (e) Outputting the recognition result to the outside of the device. and a means for doing so.

[Example 2.

The present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing one embodiment of the present invention.

The voice recognition device shown in the first section includes a voice input section 11, a voice output section 12. Control recognition unit 13. Storage section 14゜operation command section 15
Equipped with.

The audio input section 11 amplifies an audio signal input from a microphone, wireless input device, etc. (not shown), and converts it into numerical data using an A/D converter or the like.

The audio output unit 12 converts the standard pattern stored in the storage unit 14 into an audio signal using an A converter or the like and outputs the standard pattern to earphones, speakers, and wireless output devices (not shown).

The control recognition section 13 controls each function and compares the numerical data inputted through the voice input section 11 with a standard pattern stored in the storage section 14 to perform recognition processing.

The storage unit 14 stores nine programs, data, and standard patterns.

The operation command unit 15 is for communicating with externally connected equipment such as a keyboard display or a host computer.

When registering a standard pattern, the control recognition section 13 or the voice input section 11 sends an audio signal converted into numerical data to the storage section 14 according to a command from the operation command section 15 and a program stored in the storage section 14.
The standard pattern is stored as a standard pattern in the audio output section 12 and converted into an audio signal.

In the embodiment described above, this function is used when registering a standard pattern, but the voice input during the recognition operation may be output as a voice using this function.

Furthermore, the standard pattern audio output function may be combined with the recognition result confirmation audio response function.

〔Effect of the invention〕

As explained above, the present invention facilitates the creation of correct standard patterns by allowing the user to visually confirm whether the registered standard patterns are free from interruptions at the beginning or end of words, or if there is any interference with ambient noise. This has the effect of improving the recognition rate of a speaker-specific speech recognition device.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, and FIG. 2 is a block diagram showing a conventional example. ]1... Audio input section, 12... Audio output section, 13.
...Control recognition unit, 14...Storage unit, 15...Operation command unit, 21...Voice input unit, 22...Control recognition unit, 2
3...Storage unit, 24...Operation command unit.

Claims (1)

[Claims] 1. In a speaker-specific speech recognition device in which words to be recognized (hereinafter referred to as standard patterns) in the voice of the device user are registered in advance, an input speech signal is converted into numerical data. means for storing the numerical data as a standard pattern; means for converting the stored numerical data of the standard pattern into an audio signal; numerical data of the input audio signal and stored numerical data of the standard pattern. A speech recognition device comprising: a means for comparing and recognizing the results; and a means for outputting the recognition result to the outside of the device. 2. In a speaker-specific speech recognition method in which words to be recognized by the device user's voice (hereinafter referred to as standard patterns) are registered in advance, the procedure for converting input speech signals into numerical data and this numerical data A procedure for storing the numeric data of the memorized standard pattern as a standard pattern, a procedure for converting the numeric data of the memorized standard pattern into an audio signal, and comparing and recognizing the numeric data of the input audio signal with the numeric data of the memorized standard pattern. A speech recognition method comprising a procedure and a procedure for outputting a recognition result outside the device.