Speech recognition system with network accessible speech processing resources

• US 6,785,647 B2
• Filed: 04/20/2001
• Issued: 08/31/2004
• Est. Priority Date: 04/20/2001
• Status: Active Grant

• Current Assignee: Sensory Incorporated
• Original Assignee: William R. Hutchison
• Inventors: Hutchison, William R.
• Primary Examiner(s): MCFADDEN, SUSAN IRIS
• Application Number:
  US09/838,973
• Publication Number:
  US 20020156626A1
• Time in Patent Office: 1,229 Days
• Field of Search: 704/257, 704/235, 704/273, 704/243, 704/240, 704/246, 704/231, 704/201, 704/270, 704/270.1, 371/88.01
• US Class Current: 704/231
• CPC Class Codes: G10L 15/16 using artificial neural networks
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

A method of speech recognition including receiving speech signals into a front-end processor and storing at least some resources used for speech recognition in a network-attached server. The front-end processor is coupled to the network-attached server to perform the speech recognition.

Claims

1. A method of speech recognition comprising the acts of:

receiving speech signals into a front-end processor;

storing at least some speaker-dependent and/or speaker group-dependent resources used for speech recognition in a network-attached server, including resources that implement a mapping between the speech signals and tokens that have mean ing to a voice enabled application;

coupling the front-end processor to the network-attached server over a network to perform the speech recognition using the resources.

2. The method of claim 1 wherein the front-end processor comprises an untrained and/or minimally trained voice recognition system whereby the resources provided by the network-attached server provide training for the minimally trained voice recognition system.

3. The method of claim 1 further comprising identifying a speaker in the front-end processor.

4. The method of claim 1 wherein the stored resources comprise a speaker signature data structure comprising at least one speaker dependent voice model.

5. The method of claim 1 further comprising the acts of:

transmitting the speech signal to the network-attached server; and

performing speech recognition of the speech signal in the network-attached processor.

6. The method of claim 5 further comprising the acts of:

returning a response from the network-attached server over the network to the front-end processor comprising phoneme probabilities corresponding to the speech signal.

7. The method of claim 5 further comprising the acts of:

returning a response from the network-attached server to the front-end processor comprising text corresponding to the speech signal.

8. The method of claim 1 wherein the front-end processor comprises a trainable neural network and the stored resources comprise a neural network training file.

9. The method of claim 1 wherein the stored resources comprise voice signal:

value pairs associating a particular value with each voice signal.

10. A speech recognition server comprising:

a network interface configured to receive a request from an external entity;

an identification of a speaker associated with each request;

speaker-dependent signature data structures stored in the speech recognition server; and

means for generating a response including speaker-dependent voice recognition resources in response to the received request, wherein the voice recognition resources are used by the external entity to perform speech recognition.

11. The server of claim 10 further comprising encoded speech signals within each request.

12. The server of claim 11 wherein the means for generating a response comprises a neural network operable to receive the encoded speech signals and generate an output comprising values representing the language-based content of the encoded speech signals.

13. The server of claim 11 wherein the response includes phoneme probabilities corresponding to the speech signal.

14. The server of claim 11 wherein the response includes text corresponding to the speech signal.

15. The server of claim 10 further comprising speaker group signature data structures stored in the speech recognition server, wherein the means for generating a response includes the speaker group resources when speaker-dependent resources are not available.

16. The server of claim 10 wherein the means for generating a response generates a response including a voice model corresponding to the identified speaker.

17. The server of claim 10 further comprising:

an interface for receiving a feedback message corresponding to a response, the feedback message indicating whether the voice recognition resources supplied in the response were useful; and

processes executing within the server for adapting the speaker dependent signature data structures in response to the feedback message.

18. A speech recognition system comprising:

a centralized resource of shared, speaker-dependent speech recognition resources;

two or more applications having processes for receiving a voice signal and communicating with the centralized resource over a network to perform speech recognition on the voice signal using the speech recognition resources from the centralized resource.

19. The speech recognition system of claim 18 wherein the speaker-dependent speech recognition resources comprise voice models of individual speakers.

20. The speech recognition system of claim 18 wherein the speaker-dependent speech recognition resources comprise voice models of groups of speakers.

21. The speech recognition system of claim 18 wherein the speaker-dependent speech recognition resources comprise feature extraction processes.

22. The speech recognition system of claim 21 wherein the speaker-dependent speech recognition resources comprise processes operable to associative linguistic tokens with the extracted features.

23. The speech recognition system of claim 18 wherein the speaker-dependent speech recognition resources comprise speech samples for individual speakers.

24. The speech recognition system of claim 18 further comprising:

a feedback message generated by one of the applications to the centralized resource, the feedback message indicating efficacy of the shared speech recognition resource; and

processes within the centralized resource operable to use the feedback message to adapt the shared speech recognition resource to improve efficacy.

25. A speech-enabled software application comprising:

a first interface for receiving a voice signal from a speaker;

a second interface for sending the voice signal over a network to a centralized speech recognition server;

a third interface for receiving phoneme probabilities from the speech recognition server corresponding to the voice signal; and

processes for using the phoneme probabilities to launch speech-enabled functions for the speaker.

26. A method of creating a speech sample database comprising:

accepting a voice recognition task at an application;

communicating the voice recognition task to a centralized resource;

performing the voice recognition task at a the centralized resource;

causing the application to evaluate correctness of the voice recognition; and

storing the speech sample from the task with its recognition result in a speech sample database.

27. The method of claim 26 further comprising storing context information metadata describing the speaker with an association to the sample.

28. The method of claim 26 further comprising providing speech samples meeting specified criteria to external entities.