System and method for speech recognition system

• US 8,099,284 B2
• Filed: 05/03/2010
• Issued: 01/17/2012
• Est. Priority Date: 04/03/2003
• Status: Expired due to Fees

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: Arizmendi, Iker, Parthasarathy, Sarangarajan, Rose, Richard Cameron
• Primary Examiner(s): Armstrong, Angela A
• Application Number:
  US12/772,564
• Publication Number:
  US 20100211396A1
• Time in Patent Office: 624 Days
• Field of Search: 704/201, 704/231, 704/251, 704/270, 704/270.1
• US Class Current: 704/270.1
• CPC Class Codes: G10L 15/285 Memory allocation or algorithm optimisation to reduce hardware requirements
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications

Abstract

A digital speech enabled middleware module is disclosed that facilitates interaction between a large number of client devices and network-based automatic speech recognition (ASR) resources. The module buffers feature vectors associated with speech received from the client devices when the number of client devices is greater than the available ASR resources. When an ASR decoder becomes available, the module transmits the feature vectors to the ASR decoder and a recognition result is returned.

Claims

1. A method of processing speech data, the method comprising, at a speech server:

receiving an audio segment;

acquiring a decoder proxy for the audio segment from a decoder proxy cache;

passing to the decoder proxy a recognized phrase generated by performing speech recognition on the audio segment; and

issuing a query based on the recognized phrase.

2. The method of claim 1, wherein the speech server receives the speech request using a custom protocol.

3. The method of claim 1, wherein a speech server dispatcher notifies a session object of the audio segment.

4. The method of claim 1, wherein a session object further analyzes the audio segment to determine a type of application-specific handler to generate.

5. The method of claim 4, wherein a session object further stores client states for a duration of a user device session.

6. The method of claim 5, wherein the session object further stores transient acoustic information.

7. The method of claim 1, further comprising, if no recognized phrase associated with the result is identified, communicating a failure code to the user device.

8. A method of processing speech data, the method comprising, at a speech sensor:

receiving an audio segment;

processing the audio segment via a decoder proxy acquired from a decoder proxy cache;

if the decoder proxy is successfully acquired, streaming computed cepstrum vectors to a decoding process; and

if the decoder proxy is not successfully acquired, buffering the computed cepstrum vectors and transmitting the computed cepstrum vectors when the decoder proxy is acquired.

9. The method of claim 8, further comprising, after processing the audio segment via the decoder proxy, returning control to a server dispatcher for servicing other user devices.

10. The method of claim 9, wherein the server dispatcher manages speech requests from multiple user devices on a speech segment by speech segment basis.

11. The method of claim 9, further comprising:

notifying the server dispatcher of a recognition result from the decoder process;

transmitting a decoder result to the decoder proxy;

transmitting a recognized phrase from the decoded result from the decoder proxy to a handler; and

using the recognized phrase to query an application server.

12. The method of claim 8, further comprising:

applying at least one acoustic compensation algorithm to the audio segment.

13. The method of claim 12, wherein the at least one acoustic compensation algorithm comprises frequency warping-based speaker normalization.

14. The method of claim 12, wherein the at least one acoustic compensation algorithm comprises constrained model adaptation.

15. The method of claim 12, wherein the at least one acoustic compensation algorithm comprises speaker adaptive training.

16. The method of claim 12, wherein the at least one acoustic compensation algorithm comprises cepstrum and variance normalization.

17. A system for managing speech segments received from a plurality of user devices, the system comprising:

a processor;

a server dispatcher module for controlling the processor to detect and route system I/O events;

a decoder proxy cache containing a plurality of decoder proxies, each decoder proxy representing a decoder process;

a handler module controlling the processor to acquire a decoder proxy for a speech segment from the decoder proxy cache; and

a decoder processing module for controlling the processor to receive the speech segment from the decoder proxy and return the automatic speech recognition result to the decoder proxy, wherein the handler module further controls the processor to query an application sensor based on the automatic speech recognition result.

18. The system for managing speech segments of claim 17, wherein the handler module further controls the processor to perform cepstrum feature analysis on the speech segment.

19. The system for managing speech segments of claim 18, wherein the handler module further controls the processor to perform acoustic feature space normalizations and transformations on the speech segment.

20. The system for managing speech segments of claim 19, wherein parameters used for the acoustic feature space normalizations and transformations on the speech segment are estimated using speech obtained from previous utterances spoken by the user of the user device.