SYSTEM AND METHOD FOR PERSONALIZATION OF ACOUSTIC MODELS FOR AUTOMATIC SPEECH RECOGNITION

• US 20110066433A1
• Filed: 09/16/2009
• Published: 03/17/2011
• Est. Priority Date: 09/16/2009
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: CASEIRO, Diamantino Antonio, CONKIE, Alistair D., LJOLJE, Andrej
• Granted Patent: US 9,026,444 B2
• Time in Patent Office: Days
• US Class Current: 704/236
• CPC Class Codes: G10L 15/04 Segmentation; Word boundary detection
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/07 to the speaker
• CPC Class Codes: G10L 15/083 Recognition networks G10L15/142, G10L15/16 take precedence
• CPC Class Codes: G10L 15/14 using statistical models, e.g. Hidden Markov Models [HMMs] G10L15/18 takes precedence
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/28 Constructional details of speech recognition systems
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems

Abstract

Disclosed herein are methods, systems, and computer-readable storage media for automatic speech recognition. The method includes selecting a speaker independent model, and selecting a quantity of speaker dependent models, the quantity of speaker dependent models being based on available computing resources, the selected models including the speaker independent model and the quantity of speaker dependent models. The method also includes recognizing an utterance using each of the selected models in parallel, and selecting a dominant speech model from the selected models based on recognition accuracy using the group of selected models. The system includes a processor and modules configured to control the processor to perform the method. The computer-readable storage medium includes instructions for causing a computing device to perform the steps of the method.

Claims

1. A method of automatic speech recognition, the method causing a processor to perform steps comprising:

selecting a speaker independent model;

selecting a quantity of speaker dependent models, the quantity of speaker dependent models being based on available computing resources, wherein a group of selected models comprises the speaker independent model and the quantity of speaker dependent models;

recognizing an utterance using each of the models in the group of selected models in parallel; and

selecting a dominant speech model from the group of selected models based on recognition accuracy using the group of selected models.

2. The method of claim 1, further comprising dropping one of the selected speech models if a recognition efficiency is below a threshold.

3. The method of claim 1, further comprising identifying a communication device and selecting speaker dependent models based on the identified communication device.

4. The method of claim 3, further comprising selecting speaker dependent models based on a user profile associated with the communication device.

5. The method of claim 3, further comprising receiving utterances from the communication device and clustering the utterances to generate a new speaker dependent model.

6. The method of claim 1, further comprising iteratively performing the selecting, recognizing, and selecting steps each time an utterance is received until a conversation associated with the utterances terminates.

7. The method of claim 1, further comprising iteratively performing the selecting, recognizing, and selecting steps until a dominant speech model is selected.

8. A system for automatic voice recognition, the system comprising:

a processor;

a module configured to control the processor to select a speaker independent model;

a module configured to control the processor to select a quantity of speaker dependent models, the quantity of speaker dependent models being based on available computing resources, wherein a group of selected models comprises the speaker independent model and the quantity of speaker dependent models;

a module configured to control the processor to recognize an utterance using each of the models in the group of selected models in parallel;

a module configured to control the processor to select a dominant speech model from the group of selected models; and

a module configured to control the processor to recognize additional received speech using the dominant speech model.

9. The system of claim 8, further comprising a module configured to control the processor to drop one of the selected speech models if recognition efficiency is below a threshold.

10. The system of claim 8, further comprising a module configured to control the processor to identify a communication device and select speaker dependent models based on the identified communication device.

11. The system of claim 10, further comprising a module configured to control the processor to select speaker dependent models based on a user profile associated with the communication device.

12. The system of claim 10, further comprising a module configured to control the processor to receive utterances from the communication device and cluster the utterances to generate a new speaker dependent model.

13. The system of claim 8, further comprising a module configured to control the processor to iteratively perform the selecting, recognizing, and selecting steps each time an utterance is received until a conversation associated with the utterances terminates.

14. The system of claim 8, further comprising a module configured to control the processor to iteratively perform the selecting, recognizing, and selecting steps until a dominant speech model is selected.

15. A computer-readable storage medium storing a set of program instructions executable on a data processing device and usable to recognize speech, the instructions causing the data processing device to perform the steps:

identifying a communication device;

selecting a plurality of speech models based on the identified communication device;

determining a speech recognition efficiency for each of the plurality of speech models by recognizing an utterance using each of the plurality of speech models in parallel; and

if each determined speech recognition efficiency is below a threshold;

collecting the utterance;

generating a cluster of utterances; and

generating a new speech model based on the cluster of utterances.

16. The computer-readable storage medium of claim 15, wherein the instructions further comprise iteratively performing the steps of collecting and clustering for each new utterance.

17. The computer-readable storage medium of claim 15, wherein the instructions further comprise dropping speech models with speech recognition efficiencies below the threshold and selecting new speech models.

18. The computer-readable storage medium of claim 15, wherein the instructions further comprise recognizing new utterances using the new speech model.

19. The computer-readable storage medium of claim 15, wherein the instructions further comprise recognizing utterances using a speaker independent speech model while the new speech model is generated.

20. The computer-readable storage medium of claim 15, wherein the instructions further comprise identifying ambient noise associated with the utterance.