MICROPHONE-ARRAY-BASED SPEECH RECOGNITION SYSTEM AND METHOD

• US 20130030803A1
• Filed: 10/12/2011
• Published: 01/31/2013
• Est. Priority Date: 07/26/2011
• Status: Active Grant

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: Liao, Hsien-Cheng
• Granted Patent: US 8,744,849 B2
• Time in Patent Office: Days
• US Class Current: 704/233
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 2021/02166 Microphone arrays; Beamforming
• CPC Class Codes: G10L 21/0208 Noise filtering
• CPC Class Codes: H04R 1/406 microphones
• CPC Class Codes: H04R 3/005 for combining the signals of two or more microphones specially adapted for hearing aids H04R25/407

Abstract

A microphone-array-based speech recognition system combines a noise cancelling technique for cancelling noise of input speech signals from an array of microphones, according to at least an inputted threshold. The system receives noise-cancelled speech signals outputted by a noise masking module through at least a speech model and at least a filler model, then computes a confidence measure score with the at least a speech model and the at least a filler model for each threshold and each noise-cancelled speech signal, and adjusts the threshold to continue the noise cancelling for achieving a maximum confidence measure score, thereby outputting a speech recognition result related to the maximum confidence measure score.

Claims

1. A microphone-array-based speech recognition system, combining a noise masking module for cancelling noise of input speech signals from an array of microphones, according to an inputted threshold, and comprising:

at least a speech model and at least a filler model that receive respectively a noise-cancelled speech signal outputted by said noise masking module;

a confidence measure score computation module that computes a confidence measure score with said at least a speech model and said at least a filler model for said threshold and the noise-cancelled speech signal; and

a threshold adjustment module that adjusts said threshold and provides to said noise masking module to continue the noise cancelling for achieving a maximum confidence measure score computed by said confidence measure score computation module, thereby outputting a speech recognition result related to said maximum confidence measure score.

2. The speech recognition system as claimed in claim 1, wherein said threshold adjustment module finds a threshold corresponding to said maximum confidence measure score by applying an expectation-maximization algorithm.

3. The speech recognition system as claimed in claim 1, said system compares the similarity of said noise-cancelled speech signal with each of said at least a speech model and obtains a score for each said speech model, and compares the similarity of said noise-cancelled speech signal with at least a filler model and obtains a score for said at least a filler model, and said confidence measure score computation module substrates the score for said at least a filler model from a score function value for said at least a speech model, and takes the difference as said confidence measure score.

4. The speech recognition system as claimed in claim 2, wherein said at least a speech model includes N speech models, said threshold adjustment module takes the M highest scores of M speech models among said N speech models and then gives each of said M speech models different weights to find the threshold corresponding to said maximum confidence measure score, wherein N and M are all positive integers and M≦

N.

5. The speech recognition system as claimed in claim 2, wherein said at least a speech model includes a plurality of speech models, and said threshold adjustment module takes a score of a combined speech model merged from each of said a plurality of speech models to find the threshold corresponding to said maximum confidence measure score.

6. The speech recognition system as claimed in claim 2, wherein said at least a speech model includes a plurality of speech models, and said threshold adjustment module takes the maximum score for each of said a plurality of speech models to find the threshold corresponding to said maximum confidence measure score.

7. The speech recognition system as claimed in claim 1, said speech recognition system further comprises at least a processor to complete functional implementation of said at least a speech model, said at least a filler model, said confidence measure score computation module, and said threshold adjustment module.

8. The speech recognition system as claimed in claim 1, wherein said at least a speech model, said at least a filler model, said confidence measure score computation module, and said threshold adjustment module are realized by at least an integrated circuits.

9. A microphone-array-based speech recognition system combining a noise masking module for cancelling noise of input speech signals from an array of microphones, according to each of a plurality of given thresholds within a predetermined range, and comprising:

at least a speech model and at least a filler model that receive respectively a noise-cancelled speech signals after said cancelling noise;

a confidence measure score computation module that computes a confidence measure score with said at least a speech model and said at least a filler model for each given threshold within said predetermined range and said noise-cancelled speech signals; and

a maximum confidence measure score decision module that determines a maximum confidence measure score from all confidence measure scores computed by said confidence measure score computation module and obtains a threshold corresponding to said maximum confidence measure score, and outputs corresponding speech recognition result.

10. The speech recognition system as claimed in claim 9, said speech recognition system further comprises at least a processor to complete functional implementation of said at least a speech model, said at least a filler model, said confidence measure score computation module, and said maximum confidence measure score decision module.

11. The speech recognition system as claimed in claim 9, wherein said at least a speech model, said at least a filler model, said confidence measure score computation module, and said maximum confidence measure score decision module are realized by at least an integrated circuits.

12. The speech recognition system as claimed in claim 9, said speech recognition system finds the threshold corresponding to said maximum confidence measure score by applying a linear search method.

13. A microphone-array-based speech recognition method implemented by a computer system, said method comprising following acts executed by said computer system:

executing noise cancelling of input speech signals from an array of microphones according to at least an inputted threshold, and transmitting a noise-cancelled speech signal to at least a speech model and at least a filler model respectively;

computing a corresponding confidence measure score based on score information for each of said at least a speech model and a score for said at least a filler model;

from each of said at least an inputted threshold, finding a threshold corresponding to a maximum confidence measure score among all computed confidence measure scores, and generating speech recognition result.

14. The speech recognition method as claimed in claim 13, said finds the threshold corresponding to said maximum confidence measure score by applying an expectation-maximization algorithm.

15. The speech recognition method as claimed in claim 13, said method finds the threshold corresponding to said maximum confidence measure score by applying a linear search scheme.

16. The speech recognition method as claimed in claim 13, said method substrates said score for said at least a filler model from each score function value for said at least a speech model for each of said at least an inputted threshold, and takes obtained difference as each corresponding confidence measure score.

17. The speech recognition method as claimed in claim 14, said method combines said at least a speech model into a combined model to increase robustness.

18. The speech recognition method as claimed in claim 14, wherein said at least a speech model includes N speech models, and said method takes M highest scores of M speech models among said N speech models and then gives each of said M speech models different weights to find the threshold corresponding to said maximum confidence measure score, wherein N and M are all positive integers and M≦

N.

19. The speech recognition method as claimed in claim 14, wherein said at least a speech model comprises a plurality of speech models, and said method takes the maximum confidence measure score for each of said a plurality of speech models as a score function value for said at least a speech model.

20. The speech recognition method as claimed in claim 15, said method combines said at least a speech model into a combined model to increase robustness.