Frame erasure concealment technique for a bitstream-based feature extractor

• US 10,109,271 B2
• Filed: 05/19/2014
• Issued: 10/23/2018
• Est. Priority Date: 12/10/1999
• Status: Expired due to Term

• Current Assignee: Cerence Operating Company (Cerence Inc.)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Cox, Richard Vandervoort, Kim, Hong Kook
• Primary Examiner(s): Opsasnick, Michael N
• Application Number:
  US14/281,026
• Publication Number:
  US 20140330564A1
• Time in Patent Office: 1,618 Days
• Field of Search: 704236
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 19/005 Correction of errors induced by the transmission channel, if related to the coding algorithm

Abstract

A frame erasure concealment technique for a bitstream-based feature extractors in a speech recognition system particularly suited for use in a wireless communication system operates to “delete” each frame in which an erasure is declared. The deletions thus reduce the length of the observation sequence, but have been found to provide for sufficient speech recognition based on both single word and “string” tests of the deletion technique.

Claims

1. A method comprising:

receiving, at a speech processing system, a speech signal representing audible speech;

measuring, via a processor of the speech processing system, a distance between first line spectrum pair coefficients of a first frame and second line spectrum pair coefficients of a second frame representative of the speech signal in a bitstream, wherein the first frame and the second frame are adjacent frames;

comparing, via the processor of the speech processing system, the distance to a steady-state threshold distance, wherein the steady-state threshold distance identifies a steady-state region;

when the distance is one of less than or equal to the steady-state threshold distance, deleting one of the first line spectrum pair coefficients and the second line spectrum pair coefficients from the bitstream to yield a modified bitstream that represents the speech signal; and

performing automatic speech recognition of the speech signal by processing, via the processor of the speech processing system, the modified bitstream.

2. The method of claim 1, wherein the distance is further measured using an empirical mean value of the first line spectrum pair coefficients and the second line spectrum pair coefficients.

3. The method of claim 2, wherein the distance is further measured using a forgetting value applied to a final calculation result.

4. The method of claim 1, further comprising:

computing a probability of an observation sequence according to a decoding algorithm in response to detecting the frame is deleted in a particular speech signal.

5. The method of claim 4, further comprising:

decoding the observation sequence according to a hidden Markov model process.

6. The method of claim 5, wherein deleting the first line spectrum pair coefficients further comprises:

determining an error in bits most sensitive to error within a plurality of frames; and

declaring a frame erasure associated with the first line spectrum pair coefficients.

7. The method of claim 6, wherein the bits most sensitive to error in the plurality of frames comprises line spectrum pair information bits and gain information bits.

8. A speech processing system comprising:

a processor; and

a computer-readable storage medium having instructions stored which, when executed by the processor, cause the processor to perform operations comprising;

receiving a speech signal representing audible speech;

measuring a distance between first line spectrum pair coefficients of a first frame and second line spectrum pair coefficients of a second frame representative of the speech signal in a bitstream, wherein the first frame and the second frame are adjacent frames;

comparing the distance to a steady-state threshold distance, wherein the steady-state threshold distance identifies a steady-state region;

when the distance is one of less than or equal to the steady-state threshold distance, deleting one of the first line spectrum pair coefficients and the second line spectrum pair coefficients from the bitstream to yield a modified bitstream that represents the speech signal; and

performing automatic speech recognition of the speech signal by processing the modified bitstream.

9. The speech processing system of claim 8, wherein the distance is further measured using an empirical mean value of the first line spectrum pair coefficients and the second line spectrum pair coefficients.

10. The speech processing system of claim 9, wherein the distance is further measured using a forgetting value applied to a final calculation result.

11. The speech processing system of claim 8, the computer-readable storage medium having additional instructions stored which, when executed by the processor, result in operations comprising:

computing a probability of an observation sequence according to a decoding algorithm in response to detecting the frame is deleted in a particular speech signal.

12. The speech processing system of claim 11, the computer-readable storage medium having additional instructions stored which, when executed by the processor, result in operations further comprising:

decoding the observation sequence according to a hidden Markov model process.

13. The speech processing system of claim 12, wherein deleting the first line spectrum pair coefficients further comprises:

determining an error in bits most sensitive to error within a plurality of frames; and

declaring a frame erasure associated with the first line spectrum pair coefficients.

14. The speech processing system of claim 13, wherein the bits most sensitive to error in the plurality of frames comprises line spectrum pair information bits and gain information bits.

15. A computer-readable storage device having instructions stored which, when executed by a speech processing computing device, cause the speech processing computing device to perform operations comprising:

receiving a speech signal representing audible speech;

measuring a distance between first line spectrum pair coefficients of a first frame and second line spectrum pair coefficients of a second frame representative of the speech signal in a bitstream, wherein the first frame and the second frame are adjacent frames;

comparing the distance to a steady-state threshold distance, wherein the steady-state threshold distance identifies a steady-state region;

when the distance is one of less than or equal to the steady-state threshold distance, deleting one of the first line spectrum pair coefficients and the second line spectrum pair coefficients from the bitstream to yield a modified bitstream that represents the speech signal; and

performing automatic speech recognition of the speech signal by processing the modified bitstream.

16. The computer-readable storage device of claim 15, wherein the distance is further measured using an empirical mean value of the first line spectrum pair coefficients and the second line spectrum pair coefficients.

17. The computer-readable storage device of claim 15, wherein the distance is further measured using a forgetting value applied to a final calculation result.