Speech recognition in noisy environments

• US 6,850,887 B2
• Filed: 02/28/2001
• Issued: 02/01/2005
• Est. Priority Date: 02/28/2001
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Marcheret, Etienne, Epstein, Edward, Lewis, Burn
• Primary Examiner(s): McFadden, Susan
• Application Number:
  US09/796,322
• Publication Number:
  US 20020120443A1
• Time in Patent Office: 1,434 Days
• Field of Search: 704/231, 704/243, 704/251, 704/252, 704/255, 704/219, 704/220, 704/264, 704/262
• US Class Current: 704/252
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 2015/228 of application context

Abstract

Methods and apparatus for providing speech recognition in noisy environments. An energy level associated with audio input is ascertained, and a decision is rendered on whether to accept the at least one word as valid speech input, based on the ascertained energy level.

Claims

1. A method of providing speech recognition in noisy environments, said method comprising the steps of:

accepting audio input;

determining at least one word associated with the audio input;

ascertaining an energy level associated with the audio input;

normalizing the ascertained energy level;

ascertaining a detailed match score associated with the at least one word; and

rendering a decision on whether to accept the at least one word as valid speech input, based on the normalized energy level and match score.

2. The method according to claim 1, further comprising:

providing a graph of detailed match score versus energy;

said rendering step further comprising comparing the detailed match score to the graph.

3. The method according to claim 2, wherein said rendering step comprises accepting the at least one word as valid speech input if the detailed match score lies above the graph.

4. The method according to claim 2, further comprising determining whether the at least one word is in a natural firmed-up state.

5. The method according to claim 4, wherein rendering step comprises rejecting the at least one word as valid speech input if the detailed match score lies below the graph and the at least one word is determined as not being in a natural firmed-up state.

6. The method according to claim 4, wherein rendering step comprises holding the at least one word for further evaluation if the detailed match score lies below the graph and the at least one word is determined as being in a natural firmed-up state.

7. An apparatus for providing speech recognition in noisy environments, said apparatus comprising:

an input medium which accepts audio input;

an arrangement which determines at least one word associated with the audio input;

an arrangement which ascertains an energy level associated with the audio input;

an arrangement which norrmalizes the ascertained energy level;

an arrangement which ascertains a detailed match score associated with the at least one word; and

a decision arrangement which renders a decision on whether to accept the at least one word as valid speech input, based on the normalized energy level and match score.

8. The apparatus according to claim 7, further comprising:

a graph of detailed match score versus energy;

said decision arrangement being adapted to compare the detailed match score to the graph.

9. The apparatus according to claim 8, wherein said decision arrangement is adapted to accept the at least one word as valid speech input if the detailed match score lies above the graph.

10. The apparatus according to claim 8, further comprising an arrangement which determines whether the at least one word is in a natural firmed-up state.

11. The apparatus according to claim 10, wherein the decision arrangement is adapted to reject the at least one word as valid speech input if detailed match score lies below the graph and the at least one word is determined as not being in a natural firmed-up state.

12. The apparatus according to claim 10, wherein said decision arrangement is adapted to hold the at least one word for further evaluation if the detailed match score lies below the graph and the at least one word is determined as being in a natural firmed-up state.

13. A program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps for providing speech recognition in noisy environments, said method comprising the steps of:

accepting audio input;

determining at least one word associated with the audio input;

ascertaining an energy level associated with the audio input;

ascertaining an energy level associated with the audio input;

normalizing the ascertained energy level;

ascertaining a detailed match score associated with the at least one word; and

rendering a decision on whether to accept the at least one word as valid speech input, based on the normalized energy level and match score.