Multi-source phoneme classification for noise-robust automatic speech recognition

• US 7,319,959 B1
• Filed: 05/14/2003
• Issued: 01/15/2008
• Est. Priority Date: 05/14/2002
• Status: Expired due to Fees

• Current Assignee: Knowles Electronics Llc (Knowles Corporation)
• Original Assignee: Audience Corporation
• Inventors: Watts, Lloyd
• Primary Examiner(s): Storm; Donald L.
• Application Number:
  US10/439,284
• Time in Patent Office: 1,707 Days
• Field of Search: None
• US Class Current: 704/254
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/16 using artificial neural networks
• 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 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 25/18 the extracted parameters being spectral information of each sub-band

Abstract

A system and method are disclosed for processing an audio signal including separating the audio signal into a plurality of streams which group sounds from a same source prior to classification and analyzing each separate stream to determine phoneme-level classification. One or more words of the audio signal may then be outputted.

Claims

1. A method of processing an audio signal comprising:

computing 600 spectral values on a logarithmic frequency scale from the audio signal;

separating the 600 spectral values into a plurality of streams which group sounds from a same source prior to classification;

analyzing each separated stream to determine phoneme-level classification; and

outputting one or more words of the audio signal.

2. The method of processing an audio signal as recited in claim 1 wherein phoneme-level classification accuracy is enhanced by providing as an input to a classifier a spectral envelope.

3. The method of processing an audio signal as recited in claim 1 wherein phoneme-level classification accuracy is enhanced by providing as an input to a classifier detected transients.

4. The method of processing an audio signal as recited in claim 1 wherein phoneme-level classification accuracy is enhanced by providing as an input to a classifier pitch and voicing information.

5. The method of processing an audio signal as recited in claim 1 further comprising normalizing for speaker characteristics prior to classification.

6. The method of processing an audio signal as recited in claim 1 further comprising performing noise-threshold tracking.

7. The method of processing an audio signal as recited in claim 6 further comprising adjusting gain and setting noise floor reference levels based on the noise-threshold tracking.

8. The method of processing an audio signal as recited in claim 1 further comprising training with a full phoneme target set.

9. The method of processing an audio signal as recited in claim 1 further comprising incorporating a model of syllabic stress.

10. The method of processing an audio signal as recited in claim 1 wherein the spectral values are computed with a 6 microsecond resolution post-interpolation.

11. The method of processing an audio signal as recited in claim 1 wherein the spectral values are updated every 22 microseconds.

12. The method of processing an audio signal as recited in claim 1 further comprising using the output for automatic voice-dialing for phones.

13. The method of processing an audio signal as recited in claim 1 further comprising using the output for automatic command of a system or device.

14. The method of processing an audio signal as recited in claim 1 further comprising using the output as an interface to a device.

15. The method of processing an audio signal as recited in claim 1 wherein the output comprises a meeting transcription.