Adaptive speech recognition method with noise compensation

• US 6,662,160 B1
• Filed: 10/26/2000
• Issued: 12/09/2003
• Est. Priority Date: 08/30/2000
• Status: Active Grant

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: Chien, Jen-Tzung, Wu, Kuo-Kuan, Chen, Po-Cheng
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): NOLAN, DANIEL A
• Application Number:
  US09/696,293
• Time in Patent Office: 1,139 Days
• Field of Search: 704/256, 704/202, 704/234, 704/251, 704/240
• US Class Current: 704/256
• 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

Abstract

An adaptive speech recognition method with noise compensation is disclosed. In speech recognition, optimal equalization factors for feature vectors of a plurality of speech frames corresponding to each probability density function in a speech model are determined based on the plurality of speech frames of the input speech and the speech model. The parameters of the speech model are adapted by the optimal equalization factor and a bias compensation vector, which is corresponding to and retrieved by the optimal equalization factor. The optimal equalization factor is provided to adjust a distance of the mean vector in the speech model. The bias compensation vector is provided to adjust a direction change of the mean vector in the speech model.

Claims

1. An adaptive speech recognition method with noise compensation capable of compensating noises of an input speech by adjusting parameters of a HMM (Hidden Markov Model) speech model, the input speech having a plurality of speech frames, the method comprising the steps of:

(A) determining, based on the plurality of speech frames of the input speech and the speech model, optimal equalization factors for feature vectors of the plurality of speech frames corresponding to each probability density function in the speech model, wherein the optimal equalization factor is determined based on the parameters θ

_ik=(μ

_ik, Σ

_ik) of the speech model, and is equivalent to a projection of the speech frame upon Σ

_ik<sup>−

1</sup>μ

_ik, where μ

_ik and Σ

_ik are respectively the mean vector and covariance matrix of the k-th mixture density function for a state ŝ

_t=i in the speech model; and

(B) adapting the parameters of the speech model by the optimal equalization factor and a bias compensation vector corresponding to and retrieved by the optimal equalization factor, wherein the optimal equalization factor is provided to adjust a distance of the mean vector in the speech model, and the bias compensation vector is provided to adjust a direction change of the mean vector in the speech model.

2. The method as claimed in claim 1, wherein the bias compensation vector is obtained and stored in a reference function table based on noisy speech data before executing a speech recognition process.

3. The method as claimed in claim 1, wherein, in step (B), the bias compensation vector is retrieved from a reference function table by using a corresponding optimal equalization factor as an index, so as to adjust the direction of the mean vector and remove projection bias.

4. The method as claimed in claim 3, wherein, the reference function table is established by the steps of:

calculating the optimal equalization factors for pairs of each speech frame and all parameters in the speech model based on the speech model and the noisy speech data.

calculating adaptation bias vectors corresponding to the optimal equalization factors; and

piecewisely estimating, based on the relation between the optimal equalization factors and the adaptation bias vectors, the bias compensation ill vectors by averaging the adaptation bias vectors.

5. The method as claimed in claim 4, wherein, the reference function table can be modified on line by actual input speech in executing a recognition process.

6. The method as claimed in claim 1, wherein, in step (B), based on the determined optimal equalization factor λ

_e and the retrieved bias compensation vector b(λ

_e), a calculation of probability for speech recognition is adapted by λ

_eμ

_ik+b(λ

_e), where μ

_ik is the mean vector of the k-th mixture density function for a state ŝ

_t=i in the speech model.

7. The method as claimed in claim 1, further comprising a step, before step (A), for performing a feature analysis on the speech frames of the input speech.

8. The method as claimed in claim 1, further comprising a step, after step (A), for executing a Viterbi decoding algorithm.

9. An adaptive speech recognition method with noise compensation capable of compensating noises of an input speech by adjusting parameters of a HMM (Hidden Markov Model) speech model, the input speech having a plurality of speech frames, the method comprising the steps of:

(A) determining, based on the plurality of speech frames of the input speech and the speech model, optimal equalization factors for feature vectors of the plurality of speech frames corresponding to each probability density function in the speech model; and

(B) adapting the parameters of the speech model by the optimal equalization factor and a bias compensation vector corresponding to and retrieved by the optimal equalization factor, wherein the optimal equalization factor is provided to adjust a distance of the mean vector in the speech model, and the bias compensation vector is provided to adjust a direction change of the mean vector in the speech model, wherein the bias compensation vector is retrieved from a reference function table by using a corresponding optimal equalization factor as an index, so as to adjust the direction of the mean vector and remove projection bias.