JPH04276799A - Speech recognition system - Google Patents

Abstract

PURPOSE:To reduce erroneous recognition due to an environmental noise in the periphery with comparatively easy processing and simple constitution. CONSTITUTION:The feature quantity of sound inputted from an ordinary microphone 1a and a skeleton conducting type microphone 1b are extracted by feature extraction parts 2a, 2b, respectively, thence, they are inputted to interval detecting parts 3a, 3b, and interval detection are performed. Interval information and feature quantity detected by the interval detecting parts 3a, 3b are inputted to a control part 4, and an interval detected as the sound by the microphone 1b out of the intervals detected by the microphone 1a is assumed as a correct sound interval. Thereby, it is possible to prevent a pattern collation from being performed by judging a tone other than the sound emitted from a user as the sound erroneously. Also, since possibility to omit the high frequency component of the sound only by the interval information of the microphone 1b exists, the interval information of the microphone 1a is used.

Description

[Detailed description of the invention]

0001

TECHNICAL FIELD The present invention relates to a speech recognition system, and more particularly to speech input means in a speech recognition system.

0002

BACKGROUND OF THE INVENTION In recent years, products that apply voice recognition have come on the market, but there are still many restrictions on how to speak, the surrounding environment, etc., and many of them are difficult to use. In particular, erroneous recognition or malfunction due to surrounding environmental noise is a big problem. For example, when used in an office, there are sounds such as operating office equipment, people's voices, and air conditioners, and when used in a car, there are engine sounds and car stereo sounds. Various countermeasures have been considered for these problems, and a relatively effective measure is to have two microphones, one for voice and one for noise, and to use the sound obtained from the voice microphone to the noise obtained from the noise microphone. There are methods such as performing recognition by subtracting components, using a highly directional microphone to pick up only the user's voice, and using a bone conduction microphone. However, each method has problems such as complicated processing, problems with the position of the microphone, and inability to pick up high-frequency components of audio, and so far there is no definitive and effective countermeasure.

0003

[Object] The present invention was made in view of the above-mentioned circumstances, and it is an object of the present invention to reduce erroneous recognition due to surrounding environmental noise with relatively easy processing and configuration.

0004

[Structure] In order to achieve the above objects, the present invention provides (1)
a first audio input means that converts sound from a microphone or the like into an electrical signal; a first feature extractor that extracts a feature of the audio input by the first audio input means; A first voice section detection section that detects a voice section from the feature extracted from the feature extraction section; and a second voice input means that detects voice using bone conduction such as a bone conduction vibration pickup. and a second feature amount extracting unit that extracts a feature amount of the voice input from the second voice input means;
A second speech section detecting section detects a speech section from the feature amount extracted from the feature amount extracting section of means for generating a pattern dictionary; a speech pattern dictionary storage section for storing the speech pattern dictionary; a pattern matching section for performing pattern matching between the speech pattern dictionary and an unknown input speech pattern; and outputting a result of the pattern matching. In a speech recognition system having a result output section, when inputting speech, only the speech detected in the first speech section detecting section that includes the speech section detected in the second speech section detecting section is regarded as the uttered speech. It is characterized by performing speech pattern dictionary registration and/or pattern matching for recognition, and furthermore, in (1) above, (
2) Even if it is detected as voice by the second voice section detection unit,
(3) canceling the voice section if it is not detected as voice by the first voice section detecting section; or (3) the level of volume detected by the first voice input section at times other than when inputting voice; is above a predetermined value, that is, when the surrounding environment noise is above a predetermined level, the speech section detected by the second section detection section is regarded as the uttered speech and the speech pattern dictionary is registered and/or recognized. The feature is that it performs pattern matching. Hereinafter, the present invention will be explained based on examples.

FIG. 1 is a block diagram for explaining one embodiment of the speech recognition system according to the present invention. In the figure, 1a is a microphone that converts a normal acoustic signal into an electrical signal, and 1b is a bone conduction type microphone. Then, feature quantities are extracted from the acoustic signals inputted from the microphones 1a and 1b and pattern matching is performed as described later.The feature quantity extraction method,
Various pattern matching methods and the like are currently being proposed, and the present invention may employ any of these methods. Furthermore, there are bone conduction type microphones 1b that are taken from the nasal bone, and those that are taken from the throat, but any type may be used.

(1) First, features of the voices inputted from the microphones 1a and 1b are extracted by the respective feature extractors 2a and 2b. The extracted feature amounts are input to the respective section detection units 3a and 3b, and section detection is performed using each algorithm. The section information and feature amounts detected by the respective section detection sections 3a and 3b are sent to the control section 4.
is input. The control unit 4 regards the section detected as voice by the microphone 1b as a correct voice section among the sections detected by the microphone 1a (see FIG. 2). This results in
To prevent sounds other than voices uttered by a user from being mistaken for voices and pattern matching performed. Furthermore, since there is a possibility that high frequency components of the voice may be lost if only the section information of the microphone 1b is used, the section information of the microphone 1a is used.

(2) Contrary to the example in (1), there is a possibility that a section may be detected as voice due to input from the microphone 1b even though no voice is input. For example, the sensor section of the bone conduction microphone 1b that is in contact with the user's body may regard a signal generated by friction with the user's body as the user moves the user's body as audio (see FIG. 3). In order to prevent this false detection, even if section detection information is generated by microphone 1b, microphone 1
If it is not detected in a, it is not considered as audio.

(3) When the surrounding environmental noise is constantly large, that is, when the audio level input from the microphone 1a always exceeds the threshold for detecting the section (see FIG. 4),
Section detection using the microphone 1a is impossible. Therefore, in this case, section detection is performed only using section detection information from the microphone 1b.

[0009]

[Effects] As is clear from the above explanation, according to the present invention, it is possible to correctly detect a section depending on whether the surrounding environmental noise is large or small.

[Brief explanation of the drawing]

FIG. 1 is a block diagram for explaining an embodiment of a speech recognition system according to the present invention.

FIG. 2 is a diagram for explaining an example of the invention according to claim 1, and is a diagram showing an example of output levels of microphones 1a and 1b.

FIG. 3 is a diagram for explaining an example of the invention according to claim 2, and is a diagram showing another example of the output levels of the microphones 1a and 1b.

FIG. 4 is a diagram for explaining an example of the invention according to claim 3, and is a diagram showing still another example of the output levels of the microphones 1a and 1b.

[Explanation of symbols]

1a, 1b...Microphone, 2a, 2b...Feature extraction unit, 3a
, 3b...Speech section detection unit, 4...Control unit, 5...Pattern dictionary storage unit, 6...Pattern matching unit, 7...Result output unit, 8...
Display section.

Claims (3)

[Claims] 1. A first audio input means for converting sound from a microphone or the like into an electrical signal; and a first feature extractor for extracting a feature of the audio input by the first audio input means. , a first speech section detection section that detects a speech section from the feature extracted from the first feature extraction section;
A second voice input means that detects voice using bone conduction such as a bone conduction vibration pickup, and a second feature amount extraction that extracts a feature amount of the voice input from the second voice input means. a second speech section detection section that detects a section of speech from the feature extracted from the second feature extraction section; means for generating a speech pattern dictionary from the feature quantity; a speech pattern dictionary storage section for storing the speech pattern dictionary; a pattern matching section for pattern matching the speech pattern dictionary and an unknown input speech pattern; In a speech recognition system comprising a result output section that outputs a result of matching, only the speech detected by the first speech section detection section that includes the speech section detected by the second speech section detection section at the time of speech input. What is claimed is: 1. A speech recognition system that performs speech pattern dictionary registration and/or pattern matching for recognition by regarding speech as uttered speech. 2. In the speech recognition system according to claim 1, even if the second speech section detecting section detects the speech, if the first speech section detecting section does not detect the speech as speech, the second speech section detecting section detects the speech. A speech recognition system characterized by canceling speech sections. 3. In the speech recognition system according to claim 1, when the volume level detected from the first speech input section is equal to or higher than a predetermined value at times other than during speech input, the second section A speech recognition system characterized in that a speech section detected by a detection unit is regarded as uttered speech and registered in a speech pattern dictionary and/or pattern matching for recognition is performed.