Speech recognition system and method which permits a speaker's utterance to be recognized using a hidden markov model with subsequent calculation reduction

• US 5,649,056 A
• Filed: 02/14/1994
• Issued: 07/15/1997
• Est. Priority Date: 03/22/1991
• Status: Expired due to Term

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation), Toshiba Corporation
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Nitta, Tsuneo
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): COLLINS, ALPHONSO
• Application Number:
  US08/195,845
• Time in Patent Office: 1,247 Days
• Field of Search: 395/2.28, 395/2.31, 395/2, 395/2.65, 395/2.48, 395/2.52, 381/43-51
• US Class Current: 704/256
• CPC Class Codes: G09B 19/04 Speaking with audible presentation of the material to be studied G09B5/04
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/144 Training of HMMs
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 25/12 the extracted parameters being prediction coefficients

Abstract

A sound analyzer sound analyzes an input speech signal to obtain feature vectors. A matrix quantizer performs a matrix quantization process between the feature vectors obtained by the sound analyzer and a phonetic segment dictionary prepared in phonetic segment units to obtain a phonetic segment similarity sequence. A PS-phoneme integrating section integrates the phonetic segment similarity sequence into a phonemic feature vector. A HMM recognizer checks the phonemic feature vector using a HMM prepared in certain units, to thereby perform a recognition process.

Claims

1. A speech recognition system comprising:

sound analyzing means for sound analyzing an input speech signal to obtain a plurality of two-dimensional feature parameters;

a phonetic segment dictionary for storing a plurality of types of phonetic segments including continuant segments, consonant segments, and boundary segments;

means, coupled to the sound analyzing means and the phonetic segment dictionary, for matrix-quantizing the plurality of two-dimensional features parameters obtained by the sound analyzing means using the plurality of types of phonetic segments stored in the phonetic segment dictionary to obtain a phonetic segment similarity vector sequence;

means for generating a phonemic feature vector sequence, wherein the phonemic feature vector sequence is comprised of elements, wherein each element represents similarity of the input to a stored phonetic segment;

verifying means, coupled to the generating means, for verifying the phonemic feature vector sequence obtained by the generating means by comparing the phonemic feature vector sequence with a previously stored continuous hidden markov model (HHM); and

means for recognizing speech based on a result of verifying the phonemic feature vector sequence.

2. The system according to claim 1, wherein the generating means comprises a plurality of maximum-value filters for generating the phonetic segment similarity vector sequence.

3. The system according to claim 1, wherein the generating means comprises a plurality of neural networks for generating the phonetic segment similarity vector sequence.

5. The system according to claim 1, wherein the means for matrix-quantizing comprises a matrix-quantizer for obtaining the phonetic segment similarity vector sequence by matrix-quantizing the plurality of two-dimensional feature parameters.

6. The system according to claim 1, wherein the means for matrix-quantizing comprises a neural network for obtaining the phonetic segment similarity vector sequence by matrix-quantizing the plurality of two-dimensional feature parameters.

4. A speech recognition method comprising the computser steps of:

sound analyzing an input speech signal to obtain a plurality of two-dimensional feature parameters;

matrix-quantizing the feature parameters obtained by the sound analyzing step using a phonetic segment dictionary storing a plurality of types of phonetic segments including continuant segments, consonant segments and boundary segments, to obtain a phonetic segment similarity vector sequence;

generating a phonemic feature vector sequence from the phonetic segment similarity vector sequence;

verifying the phonemic feature vector obtained by the generating step by comparing the phonemic feature vector sequence with a previously stored continuous hidden markov model (HMM); and

recognizing speech based on a result of verifying the phonemic feature vector sequence.