Noise reduction filter

• US 5,742,694 A
• Filed: 07/12/1996
• Issued: 04/21/1998
• Est. Priority Date: 07/12/1996
• Status: Expired due to Fees

• Current Assignee: Eatwell, Graham P.
• Original Assignee: Eatwell, Graham P.
• Inventors: Eatwell, Graham P.
• Primary Examiner(s): Kuntz, Curtis
• Assistant Examiner(s): CHIN, VIVIAN C
• Application Number:
  US08/679,245
• Time in Patent Office: 648 Days
• Field of Search: 381/94.2, 381/94.1, 381/94.3, 381/73.1, 395/2.35
• US Class Current: 381/94.2
• CPC Class Codes: H04B 1/123 using adaptive balancing or compensation means adaptive filter circuits and algorithms H03H
• CPC Class Codes: H04R 1/1083 Reduction of ambient noise active noise reduction per se G10K11/175; protective devices for the ear, e.g. providing acoustic protection A61F11/06
• CPC Class Codes: H04R 2225/43 Signal processing in hearing aids to enhance the speech intelligibility

Abstract

A noise reduction filter for enhancing noisy audio signals, such as speech or music. In accordance with the invention, a noisy signal is passed through a first adaptive prediction filter so as to obtain a first signal component corresponding to the predictable part of the noisy signal and a second signal component corresponding to a prediction error. The first and second signal components are each attenuated respectively, according to the levels of signal and noise in each component, and then recombined to form an enhanced output signal.

Claims

1. A method for enhancing a noisy signal, said noisy signal including a noise element and a signal element, comprising the steps of

passing said noisy signal through a first adaptive prediction filter so as to obtain a first signal component corresponding to a predictable component of said noisy signal and a second signal component corresponding to a prediction error,

passing the first signal component through a first attenuator to obtain a first attenuated signal component,

passing the second signal component through a second attenuator to obtain a second attenuated signal component and

combining the first and second attenuated signal components to form an enhanced output signal.

2. A method as in claim 1 in which the gain of second attenuator is adjusted according to estimates of the noise level in said noisy signal and the level of said second signal component and in which the first attenuator has a fixed gain.

3. A method as in claim 1 in which the gain of the first attenuator is adjusted according to an estimate of the noise level in said first signal component and the level of the first signal component, and the gain of the of second attenuator is adjusted according to an estimate of the noise level in said second signal component and the level of the second signal component.

4. A method as in claim 3 in which the estimates of the noise levels in said first and second signal components are determined from estimates of the characteristics of said noise element and from the coefficients of said first adaptive prediction filter.

5. A method as in claim 3 in which the gains of said first and second attenuators are further responsive to previous levels of the first and second attenuated signals.

6. A method as in claim 1 in which the first adaptive prediction filter is adapted via a gradient descent algorithm.

7. A method as in claim 1 in which the first adaptive prediction filter includes a moving average filter.

8. A method as in claim 1 in which the first adaptive prediction filter includes a modified auto-regressive moving average filter, which is responsive to a previous value of the enhanced output signal.

9. A method as in claim 1 in which the first adaptive prediction filter includes a modified auto-regressive moving average filter, which is responsive to a previous value of the attenuated first signal component.

10. A method as in claim 1 in which the first adaptive prediction filter includes a lattice filter.

11. A method as in claim 1 in which the first adaptive prediction filter includes a non-linear filter such as an artificial neural network.

12. A method for enhancing a noisy signal, said noisy signal including a noise element and a signal element, comprising the steps of:

passing said noisy signal through a whitening filter so as to whiten said noise element,

passing the output of said whitening filter through a first adaptive prediction filter so as to obtain a first signal component corresponding to the predictable component of said noisy signal and a second signal component corresponding to a prediction error,

passing said first signal component through a first attenuator to obtain a first attenuated signal component,

passing said second signal component through a second attenuator to obtain a second attenuated signal component,

combining said first and second attenuated signal components to form a first output signal, and

passing said first output signal through an inverse whitening filter to obtain an enhanced output signal,

wherein said whitening filter is adapted dining pauses in said signal element and the coefficients of the inverse whitening filter are determined by the coefficients of the whitening filter.

13. A method for enhancing a noisy signal, said noisy signal including a noise element and a signal element, comprising the steps of:

subtracting from said noisy signal an estimate of the predictable part of said noise element so as to obtain a modified noisy signal,

passing said modified noisy signal through a first adaptive prediction filter so as to obtain a first signal component corresponding to the predictable component of said noisy signal and a second signal component corresponding to a prediction error,

passing said first signal component through a first attenuator to obtain a first attenuated signal component,

passing said second signal component through a second attenuator to obtain a second attenuated signal component,

combining said first and second attenuated signal components to form a first output signal, and

passing said first output signal through an inverse whitening filter to obtain an enhanced output signal.

14. A method as in claim 13 in which the predictable part of said noise component is obtained by

subtracting a previous enhanced output signal from a previous sample of the noisy signal so as to obtain an estimate of the previous noise component, and

passing said estimate of the previous noise component through a prediction filter.

15. A noise reduction filter for enhancing a noisy signal, said noisy signal including a noise element and a signal element, said filter comprising:

a first adaptive prediction filter which is responsive to said noisy signal and produces as outputs a first signal component corresponding to a predictable component and a second signal component corresponding to a prediction error,

a first attenuator acting on said first signal component to produce a first attenuated signal component,

a second attenuator acting on said second signal component to produce a second attenuated signal component, and

a signal combiner for combining the first and second attenuated signal components to form an enhanced output signal.

16. A noise reduction filter as in claim 15, and further including gain adjustment means which adjusts the gain of said second attenuator according to estimates of the noise level in said second signal component and the level of said second signal component.

17. A noise reduction filter as in claim 15, which further includes

first gain adjustment means which adjusts the gain of the first attenuator according to estimates of the noise level in said first signal component and the level of the first signal component and

second gain adjustment means which adjusts the gain of the of second attenuator according to an estimate of the noise level in said second signal component and the level of said second signal component.

18. A noise reduction filter in claim 17 in which the estimates of the noise levels in said first and second signal components are determined from estimates of the characteristics of said noise element and from the coefficients of said first prediction filter.

19. A noise reduction filter as in claim 17 in which the gains of said first and second attenuators also depend upon previous levels of the first and second attenuated signals.

20. A noise reduction filter as in claim 15 which includes

filter adaptation means which adapts the coefficients of said first adaptive prediction filter according to a gradient descent algorithm so that the filter operates to predict the signal element of said noisy signal.

21. A noise reduction filter as in claim 15 in which the first prediction filter includes a moving average filter.

22. A noise reduction filter as in claim 15 in which the first prediction filter includes a modified auto-regressive moving average filter which is responsive to said enhanced output signal.

23. A noise reduction filter as in claim 15 in which the first prediction filter includes a modified auto-regressive moving average filter which is responsive to said attenuated first signal component.

24. A noise reduction filter as in claim 15 in which the first prediction filter includes a lattice filter.

25. A noise reduction filter as in claim 15 in which the first prediction filter includes a non-linear filter such as an artificial neural network.

26. A noise reduction filter for enhancing a noisy signal, said noisy signal including a noise element and a signal element, said filter comprising:

a whitening filter which operates on said noisy signal to produce a modified noisy signal which contain white noise,

a first adaptive prediction filter which is responsive to said modified noisy signal and produces as outputs a first signal component corresponding to a predictable component and a second signal component corresponding to a prediction error,

a first attenuator acting on said first signal component to produce a first attenuated signal component,

a second attenuator acting on said second signal component to produce a second attenuated signal component,

a signal combiner for combining the first and second attenuated signal components to form a first output signal, and

an inverse whitening filter which operates on said first output signal to produce an enhanced output signal,

wherein said whitening filter is adapted during pauses in said signal element and the coefficients of the inverse whitening filter are determined by the coefficients of the whitening filter.

27. A noise reduction filter for enhancing a noisy signal, said noisy signal including a noise element and a signal element, said filter comprising:

a subtraction means for subtracting from said noisy speech an estimate of the predictable part of said noise element so as to obtain a modified noisy signal,

a first adaptive prediction filter which is responsive to said modified noisy signal and produces as outputs a first signal component corresponding to a predictable component and a second signal component corresponding to a prediction error,

a first attenuator acting on said first signal component to produce a first attenuated signal component,

a second attenuator acting on said second signal component to produce a second attenuated signal component, and

a signal combiner for combining the first and second attenuated signal components to form an enhanced output signal.

28. A noise reduction filter as in claim 27 which further includes:

first delay means to provide a previous enhanced output signal from said enhanced output signal,

second delay means to provide a previous noisy signal from said noisy signal,

second subtraction means for subtracting a previous enhanced output signal from a previous noisy signal so as to obtain an estimate of the previous noise component, and

a second adaptive prediction filter which acts on the estimate of the previous noise component to provide said estimate of the predictable part of said noise component.