System for measuring speech content in sound

• US 6,718,301 B1
• Filed: 11/11/1998
• Issued: 04/06/2004
• Est. Priority Date: 11/11/1998
• Status: Expired due to Term

• Current Assignee: Starkey Laboratories Incorporated
• Original Assignee: Starkey Laboratories Incorporated
• Inventors: Woods, William S.
• Primary Examiner(s): MCFADDEN, SUSAN IRIS
• Application Number:
  US09/189,668
• Time in Patent Office: 1,973 Days
• Field of Search: 704/215, 704/228, 704/233
• US Class Current: 704/233
• CPC Class Codes: G01R 29/26 Measuring noise figure; Measuring signal-to-noise ratio
• CPC Class Codes: G10L 25/78 Detection of presence or absence of voice signals switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10
• CPC Class Codes: H04R 25/505 using digital signal processing

Abstract

A system for measuring speech content in sound. The system determining time local speech-to-noise ratio for one or more bands. The system using a signal related to the power of the input signal or envelope and generating a time-dependent mean of the envelope and deviation of the envelope from the mean to estimate a time-dependent speech-to-noise ratio. The system providing single band or multiband speech-to-noise estimates for signal processing applications. The system realizable in analog, digital or analog and digital combinations.

Claims

1. A method, comprising:

receiving an input signal having speech content;

extracting a signal related to a time-dependent power of the input signal, the signal representative of at least a portion of the speech content of the input signal;

determining a time-dependent mean of the signal, M, in the time domain;

determining a time-dependent deviation of the signal from the mean, D, in the time domain using the signal; and

estimating a time-dependent speech-to-noise ratio from M and D.

2. The method according to claim 1, wherein the signal is a non-negative function of the input signal.

3. The method according to claim 1, wherein the estimating includes comparing M and D to a predetermined mapping of a relationship between M and D to speech-to-noise ratio to obtain an estimated speech-to-noise ratio.

4. The method according to claim 1, further comprising converting the input signal to a digital signal.

5. The method according to claim 1, wherein the signal is converted to a digital signal.

6. The method according to claim 1, comprising:

filtering the input signal within a first bandpass frequency range to produce the signal.

7. The method of claim 1, wherein the deviation is computed using the standard deviation of the signal.

8. The method of claim 1, wherein the deviation is computed using an absolute value of the signal minus the mean.

9. The method according to claim 1, wherein the time-dependent mean is extracted over a discrete time interval.

10. The method according to claim 1, wherein the time-dependent mean is extracted recursively in time.

11. The method according to claim 1, wherein the extracting comprises full wave rectification of the input signal.

12. The method according to claim 1, wherein the extracting comprises half wave rectification of the input signal.

13. The method according to claim 1, wherein the extracting comprises squaring of the input signal.

14. The method according to claim 1, wherein the extracting comprises magnitude extraction of the input signal.

15. The method according to claim 1, wherein the extracting comprises low pass filtering of a rectified, squared, or magnitude extracted version of the input signal.

16. The method according to claim 1, wherein the extracting comprises Hilbert transforming the input signal.

17. The method according to claim 1, wherein a non-negative element is used for time-dependent power extraction of the input signal.

18. The method according to claim 1, wherein the estimated speech-to-noise ratio is used to set an amplifier gain in a first bandpass frequency range.

19. The method according to claim 1, wherein the estimated speech-to-noise ratio is used to determine if speech-like sound is present in the input signal.

20. The method according to claim 1, wherein if the estimated speech-to-noise ratio is below a threshold, processing the input signal as noise.

21. The method according to claim 1, wherein if the estimated speech-to-noise ratio is below a threshold, decreasing the gain of an amplifier in a first bandpass frequency range.

22. The method according to claim 1, wherein if the estimated speech-to-noise ratio is above a threshold, processing the input signal as speech.

23. The method according to claim 1, wherein if the estimated speech-to-noise ratio is above a threshold, increasing the gain of an amplifier in a first bandpass frequency range.

24. The method according to claim 1, wherein the estimated speech-to-noise ratio is used to gradually adjust the gain applied to an amplifier in a first bandpass frequency range.

25. The method according to claim 1, wherein the estimated speech-to-noise ratio is used to ignore input signals in a first bandpass frequency range if it is determined that the input signals in the first bandpass frequency range are dominated by noise.

26. The method according to claim 1, wherein the signal is an envelope of the input signal.

27. The method according to claim 26, wherein the envelope is converted to a digital signal.

28. The method of claim 26, wherein the deviation is computed using a standard deviation of the envelope.

29. The method of claim 26, wherein the deviation is computed using an absolute value of the envelope minus the mean.

30. A method, comprising:

receiving an audio signal having speech content;

converting the audio signal to an electrical signal including the speech content;

converting the electrical signal to a digital representation;

bandpass filtering the electrical signal using one or more digital filters to produce a plurality of filtered digital signals;

for each digital signal of the plurality of filtered digital signals;

extracting an envelope related to a time-dependent power of the electrical signal, the envelope representative of at least a portion of the speech content of the electrical signal;

determining a time-dependent mean of the envelope, M, in the time domain;

determining a time-dependent deviation of the envelope from the mean, D, in the time domain using the envelope; and

estimating a time-dependent speech-to-noise ratio from M and D using a predetermined mapping or relationship between M and D to speech-to-noise ratio;

producing a processed digital signal using the plurality of filtered digital signals and their respective estimated time-dependent speech-to-noise ratios; and

converting the processed digital signal to a processed analog signal.

31. The method according to claim 30, wherein the extracting comprises low pass filtering of a rectified, squared, or magnitude extracted version of the signal.

32. The method according to claim 30, wherein the extracting comprises Hilbert transforming the signal.

33. The method according to claim 30, wherein a non-negative element is used for time-dependent power extraction of the si envelope.

34. The method according to claim 30, wherein the estimated speech-to-noise ratios are used to set amplifier gains in a plurality of amplifiers.

35. The method according to claim 30, wherein the estimated speech-to-noise ratios are used to determine if speech-like sound is present in each band of the audio signal.

36. The method according to claim 30, wherein if an estimated speech-to-noise ratio is below a threshold, processing the time-dependent signal in its respective band as noise.

37. The method according to claim 30, wherein if an estimated speech-to-noise ratio is below a threshold, decreasing a gain of an amplifier.

38. The method according to claim 30, wherein if an estimated speech-to-noise ratio is above a threshold, processing the time-dependent signal as speech.

39. The method according to claim 30, wherein if an estimated speech-to-noise ratio is above a threshold, increasing a gain of an amplifier in a first bandpass frequency range.

40. The method according to claim 31, wherein if an estimated speech-to-noise ratio is above a threshold, increasing a gain of an amplifier in a first bandpass frequency range.

41. The method according to claim 30, wherein if an estimated speech-to-noise ratio is below a threshold, ignoring the time-dependent signal in its respective band as noise.