APPLYING SPEECH RECOGNITION ADAPTATION IN AN AUTOMATED SPEECH RECOGNITION SYSTEM OF A TELEMATICS-EQUIPPED VEHICLE

• US 20080010057A1
• Filed: 07/05/2006
• Published: 01/10/2008
• Est. Priority Date: 07/05/2006
• Status: Active Grant

• Current Assignee: General Motors LLC (General Motors Company)
• Original Assignee: General Motors Corporation (Motors Liquidation Co. GUC Trust)
• Inventors: Chengalvarayan, Rathinavelu, Correia, John J, Pennock, Scott M
• Granted Patent: US 7,725,316 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 2015/0635 updating or merging of old and new templates; Mean values; Weighting

Abstract

A speech recognition adaptation method for a vehicle having a telematics unit with an embedded speech recognition system. Speech is received and pre-processed to generate acoustic feature vectors, and an adaptation parameter is applied to the acoustic feature vectors to yield transformed acoustic feature vectors. The transformed acoustic feature vectors are decoded and a hypothesis of the speech is selected, and the adaptation parameter is trained using acoustic feature vectors from the hypothesis. The method also includes one or more of the following steps: the speech is observed for a certain characteristic and the trained adaptation parameter is saved in accordance with the certain characteristic for use in transforming feature vectors of subsequent speech having the certain characteristic; use of the trained adaptation parameter persists from one vehicle ignition cycle to the next; and use of the trained adaptation parameter is ceased upon detection of a system fault.

Claims

1. A method of speech recognition adaptation for a vehicle having a telematics unit with an embedded speech recognition system, the method comprising the steps of:

(a) receiving speech;

(b) pre-processing a segment of the speech to generate acoustic feature vectors;

(c) applying at least one adaptation parameter to the acoustic feature vectors to yield transformed acoustic feature vectors;

(d) decoding the transformed acoustic feature vectors to select a hypothesis therefrom corresponding to the received speech;

(e) training the at least one adaptation parameter with acoustic feature vectors of the selected hypothesis to yield at least one trained adaptation parameter; and

(f) ceasing use of the at least one adaptation parameter upon detection of a system fault.

2. The speech recognition method of claim 1, wherein the ceasing step (f) comprises:

(1) determining if the at least one trained adaptation parameter has become overtrained; and

(2) resetting the at least one trained adaptation parameter to default values if the at least one trained adaptation parameter has become overtrained.

3. The speech recognition method of claim 2, wherein the determining step includes at least one of detecting divergence of the adaptation process or a predetermined number of speech recognition failures.

4. The speech recognition method of claim 1, further comprising the steps of:

(g) pre-training at least one default adaptation parameter based on at least one of vehicle-specific or user-specific characteristics; and

(h) downloading the at least one pre-trained default adaptation parameter to the vehicle.

5. The speech recognition method of claim 4, wherein the user-specific characteristics include at least one of user gender, or user dialect based on a region where the vehicle was purchased or is registered.

6. The speech recognition method of claim 1, further comprising the step of:

(g) observing the speech for a certain characteristic and saving the at least one trained adaptation parameter in accordance with the certain characteristic for use in transforming feature vectors of subsequent speech having the certain characteristic.

7. The speech recognition method of claim 6 wherein step (g) comprises observing the speech for speech type including at least one of digit speech or command speech, saving the at least one adaptation parameter associated with the speech as a trained digit speech adaptation parameter when the speech is determined to be digit speech, and saving the at least one adaptation parameter associated with the speech as a trained command speech adaptation parameter when the speech is determined to be command speech.

8. The speech recognition method of claim 6 wherein step (g) comprises observing the speech for speech type including at least one of low noise speech or high noise speech, saving the at least one adaptation parameter associated with the speech as a trained low noise speech adaptation parameter when the speech is determined to contain low noise levels, and saving the at least one adaptation parameter associated with the speech as a trained high noise speech adaptation parameter when the speech is determined to contain high noise levels.

9. The speech recognition method of claim 6, further comprising the steps of:

(h) observing the speech for user identity and using adaptation parameters associated with the identity of the user.

10. The speech recognition method of claim 6, further comprising the steps of:

(h) using at least one of an acoustic model or a grammar developed specific to the certain characteristic.

11. The speech recognition method of claim 10, wherein step (h) comprises at least one of observing digit speech in the speech and using an acoustic model developed specific to digit speech, observing command speech in the speech and using an acoustic model developed specific to command speech, or observing a particular language of the speech and using an acoustic model developed specific to that particular language.

12. The speech recognition method of claim 10, wherein step (h) comprises observing nametag training grammar in the speech and using a grammar developed specific to nametag training.

13. The speech recognition method of claim 6, wherein the certain characteristic is a speech recognition reliability characteristic including at least one of speech length or SNR level and the method further comprises the steps of:

(h) determining whether the speech recognition reliability characteristic exceeds a minimum value; and

(i) storing the at least one trained adaptation parameter, if the speech recognition reliability characteristic exceeds the minimum value and, if not, then not storing the at least one trained adaptation parameter.

14. The speech recognition method of claim 1, further comprising the steps of:

(g) determining whether a confidence score of the selected hypothesis from the decoding step (d) exceeds a minimum confidence score; and

(h) storing the at least one trained adaptation parameter associated with the speech, if the confidence score of the selected hypothesis exceeds the minimum confidence score and, if not, then not storing the at least one trained adaptation parameter.

15. A method of speech recognition with characteristic-specific adaptation for a vehicle having a telematics unit with an embedded speech recognition system, the method comprising the steps of:

(a) receiving speech;

(b) pre-processing the speech to generate acoustic feature vectors;

(c) applying at least one adaptation parameter to the acoustic feature vectors to yield transformed acoustic feature vectors;

(d) decoding the transformed acoustic feature vectors to select a hypothesis therefrom corresponding to the received speech;

(e) training the at least one adaptation parameter with acoustic feature vectors of the selected hypothesis to yield at least one trained adaptation parameter; and

(f) observing the speech for a certain characteristic and saving the at least one trained adaptation parameter in accordance with the certain characteristic for use as at least one trained adaptation parameter in transforming feature vectors of subsequent speech having the certain characteristic.

16. The method of claim 15, wherein the certain characteristic is speech type, such that step (f) comprises observing the speech for speech type including at least one of digit speech or command speech, saving the trained adaptation parameters associated with the speech as trained digit speech adaptation parameters when the speech is determined to be digit speech, and saving the trained adaptation parameters associated with the speech as trained command speech adaptation parameters when the speech is determined to be a command speech.

17. The method of claim 16, wherein the decoding step (d) further comprises using at least one of an acoustic model or a grammar developed specific to the certain characteristic of the speech.

18. The method of claim 16, wherein step (f) comprises observing the speech for user identity and using one or more adaptation parameters associated with the identity of the user.

19. A method of speech recognition with persistence of adaptation parameters for a vehicle having a telematics unit with an embedded speech recognition system, the method comprising the steps of:

(a) receiving speech after an ignition event of the vehicle;

(b) pre-processing a segment of the speech to generate acoustic feature vectors;

(c) applying an adaptation parameter to the acoustic feature vectors to yield transformed acoustic feature vectors;

(d) decoding the transformed acoustic feature vectors to select a hypothesis therefrom corresponding to the received speech; and

(e) training the adaptation parameter with acoustic feature vectors of the selected hypothesis and saving the trained adaptation parameter for use transforming feature vectors of another segment of speech received after a subsequent ignition event.