Data processing: speech signal processing – linguistics – language – Speech signal processing – Recognition
Reexamination Certificate
1998-11-16
2001-10-09
Dorvil, Richemond (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Recognition
C704S255000
Reexamination Certificate
active
06301559
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the invention
The present invention relates to a speaker independent speech recognition method and speech recognition device, and particularly relates to a speech recognition method and speech recognition device that improve the recognizing capability for single words using a template matching system.
2. Description of the Related Art
Recently, speech recognition technology has been utilized in various fields. For example, operational control of mechanisms maintained inside a vehicle when the vehicle is traveling is carried out using sound. More specifically, operation of audio devices, an air conditioner etc., opening and closing of a window, switching on or off a light inside the vehicle, is carried out using sound. In this type of situation, it is required to reliably execute the operation of each device by correctly recognizing a sound spoken by the driver, being any unspecified speaker.
As a speech recognition system for this situation, there are various systems, one of which is the template matching system. The template matching system has a vocabulary template in which characteristic parameters for specific sounds to be recognized, for example a plurality of words, are entered in advance (stored). In a template matching system, a characteristic parameter is educed from a word spoken by an unspecified speaker, and degrees of similarity between this educed characteristic parameter and the characteristic parameters of each word entered in the vocabulary template are compared. Using this comparison, a word corresponding to a characteristic parameter among the characteristic parameters stored in the vocabulary template that is the closest approximation to the characteristic parameter of the sound spoken by the unspecified speaker is confirmed as the word spoken by the unspecified speaker. The speech waveform, frequency spectrum etc. of each word may be used as the characteristic parameter.
[Problems to be Solved by the Invention]
However, in a speech recognition method using the above described template matching system, there are the following problems.
For example, when an unspecified speaker utters, as a sound, a word that is not a word to be recognized as it is different from the specified sounds (a word that does not have a characteristic parameter entered in advance in the vocabulary template), in the above described speech recognition method, a word that is the closest approximation to the characteristic parameter of the word that is not to be recognized (one word that has a characteristic parameter entered in the vocabulary template) is erroneously recognized as the word spoken. This type of problem is particularly significant with words that are similar to specific sounds (words) to be recognized.
Also, this type of erroneous recognition is not limited to speech, but is also caused by ambient sounds that occur suddenly. In the above described speech recognition within a vehicle, ambient sounds that occur suddenly are, for example, the sound of a door opening or closing, the sound of an air conditioner operating, the sound of wind across a window while the vehicle is traveling, or the sound of tires when passing over seams or a center line etc. of a road. With these sounds as well, if the speech recognition system operates, a sound will be erroneously recognized as the word, among the words having characteristic parameters stored in the vocabulary template, having the closest approximation to the sound.
In dealing with this type of erroneous recognition, it is also preferably implemented by making the speech recognition device as large as possible, preventing increase in cost and without major changes to a conventional speech recognition device.
In order to solve the above problems, the object of the present invention is to realize a speech recognition method that limits erroneous recognition for sounds that are different from specific sounds to be recognized (for example, ambient sounds and words that are not to be recognized), and has improved recognition capability.
A further object of the present invention is to achieve the above object by making the speech recognition device as large as possible, preventing increase in cost and without major changes to a conventional speech recognition device.
SUMMARY OF THE INVENTION
In order to achieve the above described objects, a speech recognition method of the present invention has a vocabulary template with characteristic parameters for a plurality of sounds previously entered, for carrying out speech recognition for a speech that has been input using the vocabulary template, and comprises the steps of: previously entering characteristic parameters for a plurality of specific sounds to be recognized, and characteristic parameters for sounds different to the specific sounds, into the dictionary; detecting characteristic parameters from an input speech; and obtaining degrees of similarity between characteristics parameters of a detected speech and respective characteristic parameters entered in the vocabulary template, and outputting information indicating one parameter among the characteristic parameters entered in the dictionary that approximates to the educed speech as a result.
In order to achieve the above described object, a speech recognition device of the present invention for carrying out speech recognition using degrees of similarity between characteristic parameters for a plurality of previously entered speeches and characteristic parameters for input speeches, comprises: a characteristic parameter storage circuit for previously storing characteristic parameters for a plurality of specific sounds to be recognized, and characteristic parameters for sounds different to the specific sounds; a characteristic parameter educing circuit for educing characteristic parameters for input sounds; a degree of similarity comparison circuit for obtaining degrees of similarity between characteristics parameters of an educed speech and respective characteristic parameters entered in the characteristic parameter storage circuit; and a comparison result outputting circuit for outputting information indicating one degree of similarity obtained by the degree of similarity comparison circuit that approximates to the speech that has been educed among characteristic parameters stored in the characteristic parameter storage circuit.
REFERENCES:
patent: 5764852 (1998-06-01), Williams
patent: 5-066790 (1993-03-01), None
Hino Noritoshi
Shinotsuka Hiroshi
Dorvil Richemond
Frank Robert J.
OKI Electric Industry Co., Ltd.
Venable
Wood Allen
LandOfFree
Speech recognition method and speech recognition device does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Speech recognition method and speech recognition device, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Speech recognition method and speech recognition device will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2594597