Method of calculating HMM output probability and speech recognition apparatus

• US 7,058,576 B2
• Filed: 07/18/2002
• Issued: 06/06/2006
• Est. Priority Date: 07/24/2001
• Status: Expired due to Fees

• Current Assignee: Seiko Epson Corporation (Seiko Group)
• Original Assignee: Seiko Epson Corporation (Seiko Group)
• Inventors: Miyazawa, Yasunaga, Hasegawa, Hiroshi
• Primary Examiner(s): Knepper, David D.
• Application Number:
  US10/197,461
• Publication Number:
  US 20030061044A1
• Time in Patent Office: 1,419 Days
• Field of Search: None
• US Class Current: 704/256
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]

Abstract

The invention relates to speech recognition based on HMM, in which speech recognition is performed by performing vector quantization and obtaining an output probability by table reference, and the amount of computation and use of memory area are minimized while achieving a high ability of recognition. Exemplary codebooks used for vector quantization can be provided as follows: if phonemes are used as subwords, codebooks for respective phonemes, such that a codebook CB1 is a codebook for a phoneme /a/ and a codebook CB2 is a codebook for a phoneme /i/, and these codebooks are associated with respective phoneme HMMs. When a feature vector obtained by speech analysis is vector quantized based on, for example, the codebook CB1 and a code (label) is output, tables for respective states of the phoneme HMM associated with the codebook CB1 are each referred to in order to obtain state output probabilities corresponding to the label, and speech recognition is performed using the state output probabilities as a parameter.

Claims

1. A method of calculating an HMM output probability, comprising:

performing vector quantization, using vector quantization devices, of a feature vector obtained by speech analysis of an input speech; and

obtaining a state output probability in accordance with a result of quantization by referring to tables;

the vector quantization devices being associated with respective subwords, and the tables being provided in association with the vector quantization devices associated with the subwords, and

the feature vector obtained by speech analysis of the input speech being vector quantized using one of the vector quantization devices associated with subwords, a result of quantization being output, and a state output probability in accordance with the result of quantization being obtained by referring to a table associated with the vector quantization devices that has output the result of quantization.

2. The method of calculating an HMM output probability according to claim 1, the subwords being phonemes, vector quantization devices associated with phonemes being created, and the vector quantization devices associated with phonemes being associated with respective phoneme HMMs.

3. The method of calculating an HMM output probability according to claim 1, the subwords being syllables, vector quantization devices associated with syllables being created, and the vector quantization devices associated with syllables being associated with respective syllable HMMs.

4. The method of calculating an HMM output probability according to claim 1, the subwords being vowels and consonants, and the vector quantization devices associated with subwords being vector quantization devices created from vowels and vector quantization devices created from consonants.

5. The method of calculating an HMM output probability according to claim 1, of the vector quantization devices associated with subwords, vector quantization devices that can be shared, if any, being shared on the basis of each group of the respective subword HMMs.

6. The method of calculating an HMM output probability according to claim 1, of the vector quantization devices associated with subwords, vector quantization devices that can be shared, if any, being shared on the basis of each state of the respective subword HMMs.

7. The method of calculating an HMM output probability according to claim 1, each of the vector quantization devices associated with subwords having an optimal code size in accordance with the associated subword.

8. The method of calculating an HMM output probability according to claim 1, whether an input subword is a subword to be vector quantized being checked before performing vector quantization, and vector quantization being skipped if the input subword is not a subword to be vector quantized.

9. The method of calculating an HMM output probability according to claim 1, the vector quantization devices being codebooks.

10. A speech recognition apparatus for vector quantizing a feature vector obtained by speech analysis of an input speech using vector quantization devices, obtaining a state output probability in accordance with a result of quantization by referring to tables, and performing speech recognition using the state output probability as a parameter, the speech recognition apparatus comprising:

a speech analysis unit to perform speech feature analysis of an input speech to output a feature vector sequence;

a vector quantization device storage unit to store vector quantization devices associated with respective subwords;

a vector quantization processing unit to vector quantize the feature vector sequence output from the speech analysis unit using the vector quantization devices associated with subwords, stored in the vector quantization storage device storage unit, and to output a result of quantization;

a subword HMM storage unit including various parameters required for subword HMM calculation and tables to refer to state output probabilities, provided in association with the vector quantization devices associated with subwords; and

a speech recognition processing unit to refer to the table associated with vector quantization devices used for vector quantization upon receiving a result of quantization from the vector quantization devices, to obtain a state output probability in accordance with the result of quantization, and to perform speech recognition using the state output probability as a parameter to output a result of recognition.

11. The speech recognition apparatus according to claim 10, the subwords being phonemes, vector quantization devices associated with phonemes being created, and the vector quantization devices associated with phonemes being associated with respective phoneme HMMs.

12. The speech recognition apparatus according to claim 10, the subwords being syllables, vector quantization devices associated with syllables being created, and the vector quantization devices associated with syllables being associated with respective syllable HMMs.

13. The speech recognition apparatus according to claim 10, the subwords being vowels and consonants, and the vector quantization devices associated with subwords being vector quantization devices created from vowels and vector quantization devices created from consonants.

14. The speech recognition apparatus according to claim 10, of the vector quantization devices associated with subwords, vector quantization devices that can be shared, if any, being shared on the basis of each group of the respective subword HMMs.

15. The speech recognition apparatus according to claim 10, of the vector quantization devices associated with subwords, vector quantization devices that can be shared, if any, being shared on the basis of each state of the respective subword HMMs.

16. The speech recognition apparatus according to claim 10, each of the vector quantization devices associated with subwords having an optimal code size in accordance with the associated subword.

17. The speech recognition apparatus according to claim 10, a word/language table being provided, whether an input subword is a subword to be vector quantized being checked by referring to the word/language table before the vector quantization unit performs vector quantization, and vector quantization being skipped if the input subword is not a subword to be vector quantized.

18. The speech recognition apparatus according to claim 10, the vector quantization devices being codebooks.