System for separating speech from background noise

• US 5,319,736 A
• Filed: 12/06/1990
• Issued: 06/07/1994
• Est. Priority Date: 12/06/1989
• Status: Expired due to Fees

• Current Assignee: National Research Council Canada
• Original Assignee: National Research Council Canada
• Inventors: Hunt, Melvyn
• Primary Examiner(s): Knepper, David D.
• Application Number:
  US07/623,336
• Time in Patent Office: 1,279 Days
• Field of Search: 381/41-49, 381/71, 381/94, 395/2, 395/2.36, 395/2.12
• US Class Current: 704/227
• 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 2021/02165 Two microphones, one receiving mainly the noise signal and the other one mainly the speech signal
• CPC Class Codes: G10L 21/0208 Noise filtering

Abstract

A digital signal processing system provides to sequences of power spectra in each of two signal channels, one channel, termed the "speech" channel, containing speech and environmental noise and the other channel, termed the "reference" channel, containing primarily the same environmental noise. An adaptive filtering technique is employed using the power spectra in both channels when speech is not present in the speech channel to obtain a relationship between the environmental noise power spectra in the two channels. When speech is present in the speech channel, a prediction of the environmental noise power spectrum on that channel is obtained from the power spectrum of the noise on the reference channel and the relationship between the noise power spectra on the two channels previously obtained. Thus the noise power spectrum be extracted from the power spectrum of the speech and noise in the speech channel. The result is that the speech will be more easily recognizable by either human listeners or speech recognition systems.

Claims

1. An apparatus for separating speech from background noise, comprising:

means for inputting speech contaminated with background noise to provide a noisy speech signal;

means for inputting primarily the background noise contaminating the speech to provide a reference signal;

signal processing means for implementing an adaptive least-squares adaptation method to obtain an estimate of the power spectrum of the noise contaminating the speech from a plurality of recent samples of the power spectrum of the reference signal, and for subtracting said estimate of the power spectrum of the noise contaminating speech from the total power spectrum of said noisy speech signal to obtain an estimate of the power spectrum of the speech.

2. An apparatus as claimed in claim 1, in combination with a speech recognition system which receives the estimate of the power spectrum of the speech.

3. An apparatus as claimed in claim 2, wherein said signal processing means comprises means for using the Widrow-Hoff Least Mean Squares algorithm to implement said adaptive least-squares adaptation method.

4. An apparatus as claimed in claim 1, wherein said signal processing means comprises means for using the Widrow-Hoff Least Mean Squares algorithm to implement said adaptive least-squares adaptation method.

5. An apparatus as claimed in claim 1, wherein said signal processing means comprises means for combining said samples linearly using previously derived coefficients in a combining formula to implement said adaptive least-squares adaptation method, said coefficients having been deriving during recent periods when no speech was present in said noisy speech signal.

6. An apparatus as claimed in claim 5, in combination with a speech recognition system which receives the estimate of the power spectrum of the speech.

7. An apparatus as claimed in claim 5, wherein said means for combining comprises means for using the Widrow-Hoff Least Mean Squares algorithm.

8. A method of separating background noise from a noisy speech signal, comprising:

continually monitoring background noise to provide a reference signal;

processing the reference signal to obtain an estimate of the power spectrum thereof using an adaptive least-squares adaptation method from a plurality of recent samples of the power spectrum of the reference signal; and

processing the noisy speech signal by subtracting the estimate from the total power spectrum of the noisy signal to obtain an estimate of the power spectrum of the speech.

9. A method as claimed in claim 8, wherein said adaptive least-squares adaptation method uses the Widrow-Hoff Least Mean Squares algorithm.

10. A method as claimed in claim 8, wherein said adaptive least-squares adaptation method combines said samples linearly using coefficients in a combining formula that were previously derived during recent periods when no speech was present in said noisy speech signal.

11. A method as claimed in claim 10, wherein said adaptive least-squares adaptation method uses the Widrow-Hoff Least Mean Squares algorithm.

12. A method as claimed in claim 11, in combination with the step of conveying said estimate of the power spectrum of the speech to a speech recognition system.

13. A method as claimed in claim 8, in combination with the step of conveying said estimate of the power spectrum of the speech to a speech recognition system.