Robust preprocessing signal equalization system and method for normalizing to a target environment

• US 6,411,927 B1
• Filed: 09/04/1998
• Issued: 06/25/2002
• Est. Priority Date: 09/04/1998
• Status: Expired due to Fees

• Current Assignee: Panasonic Corporation Of North America (Panasonic Holdings Corporation)
• Original Assignee: Matsushita Electric Corporation Of America (Panasonic Holdings Corporation)
• Inventors: Morin, Philippe, Gelin, Philippe, Junqua, Jean-Claude
• Primary Examiner(s): Chawan, Vijay B
• Application Number:
  US09/148,401
• Time in Patent Office: 1,390 Days
• Field of Search: 704/224, 704/226, 704/227, 704/228, 704/230, 704/233, 704/248, 704/208, 704/214, 704/215, 704/253
• US Class Current: 704/224
• CPC Class Codes: G10L 15/065 Adaptation
• 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

Abstract

The audio source is spectrally shaped by filtering in the time domain to approximate or emulate a standardized or target microphone input channel. The background level is adjusted by adding noise to the time domain signal prior to the onset of speech to set a predetermined background noise level based on a predetermined target. The audio source is then monitored in real time and the signal-to-noise ratio is adjusted by adding noise to the time domain signal, in real time, to maintain a signal-to-noise ratio based on a predetermined target value. The normalized audio signal may be applied to both training speech and test speech. The resultant normalization minimizes the mismatch between training and testing and also improves other speech processing functions, such as speech endpoint detection.

Claims

1. A signal normalizer for processing an audio source comprising:

a speech signal detector receptive of said audio source for detecting when speech is present and is not present in said audio source;

a first compensation factor calculation module responsive to said speech signal detector for determining a first noise quantity and adding noise to said audio source when speech is not present in said audio source, to set the background noise level in accordance with predetermined target parameters;

a second compensation factor calculation module responsive to said speech signal detector for determining a second noise quantity for selectively adding noise to said audio source when speech is present in said audio source, to set a predetermined signal-to-noise ratio in accordance with said predetermined target parameters.

2. The signal normalizer of claim 1 further comprising a filter receptive of said audio source for spectrally shaping said audio source in accordance with predetermined target parameters.

3. The signal normalizer of claim 2 wherein said audio source is a time domain signal and wherein said filter operates on said audio source in the time domain.

4. The signal normalizer of claim 2 wherein said filter is implemented by digital process.

5. The signal normalizer of claim 2 wherein said filter is implemented by analog process.

6. The signal normalizer of claim 1 wherein said target parameters include channel parameters based on electro-acoustic properties of a predetermined audio channel.

7. The signal normalizer of claim 1 wherein said target parameters include channel parameters based on electro-acoustic properties of a predetermined microphone.

8. The signal normalizer of claim 1 wherein said target parameters include background noise value corresponding to a predetermined noise level.

9. The signal normalizer of claim 1 wherein said target parameters include signal-to-noise ratio value corresponding to a predetermined signal-to-noise ratio.

10. The signal normalizer of claim 1 wherein said speech signal detector comprises a speech endpoint detector.

11. A speech recognition system, the system comprising:

a speech recognizer of the type that is trained upon a predetermined corpus of training speech generated in a training environment and used by matching patterns in an utterance of test speech generated in a use environment; and

a normalizer for processing said training speech and said test speech by adding predetermined quantities of noise to said training speech and said test speech to minimize mismatch between said training and use environments.

12. The system of claim 11 wherein said normalizer processes said training speech and said test speech in the time domain.

13. The system of claim 11 wherein said training speech and said test speech is supplied through an audio source and wherein said normalizer comprises:

a speech signal detector receptive of said audio for detecting when speech is present and is not present in said audio source;

a first compensation factor calculation module responsive to said speech signal detector for determining a first noise quantity for applying to said audio source when speech is not present in said audio source, to establish a predetermined background noise level in accordance with predetermined target parameters;

a second compensation factor calculation module responsive to said speech signal detector for determining a second noise quantity for selectively applying to said audio source when speech is present in said audio source, to establish a predetermined signal-to-noise ratio in accordance with said predetermined target parameters.

14. The system of claim 11 wherein said normalizer modifies said training speech and said test speech such that each approach a common target characterized by predetermined target channel parameters, background noise and signal-to-noise ratio.

15. The system of claim 11 wherein said normalizer modifies said training speech based on a plurality of targets, each target characterized by predetermined target channel parameters, background noise and signal-to-noise ratio.

16. The system of claim 13, wherein said normalizer further comprises a filter receptive of said audio source for spectrally shaping said audio source in accordance with said predetermined target parameters.