SYSTEM AND METHOD FOR GENERATING A SEPARATED SIGNAL

• US 20090254338A1
• Filed: 02/27/2007
• Published: 10/08/2009
• Est. Priority Date: 03/01/2006
• Status: Active Grant

• Current Assignee: Qualcomm, Inc.
• Original Assignee: Qualcomm, Inc.
• Inventors: Visser, Erik, Chan, Kwok-Leung
• Granted Patent: US 8,898,056 B2
• Time in Patent Office: Days
• US Class Current: 704/205
• CPC Class Codes: G10L 21/0272 Voice signal separating
• CPC Class Codes: H04L 25/03159 operating in the frequency domain H04L25/03165, H04L25/03178 take precedence

Abstract

The present invention relates to blind source separation. More specifically it relates to the blind source separation using frequency domain processes.

Claims

1. A signal separation method, comprising:

receiving a set of frequency-domain first signals,

wherein each frequency-domain first signals comprises frequency-domain first signal elements corresponding to a plurality of frequencies, and

wherein the set of frequency-domain first signals comprises a plurality of signal sources;

separating the set of frequency-domain first signals into a set of frequency-domain second signals,

wherein the set of frequency-domain second signals comprises a set of separated frequency-domain second signal elements corresponding to individual frequencies of the plurality of frequencies, and

wherein each frequency-domain second signal element is assigned an identifier which indicates which of the set of frequency-domain second signals includes the frequency-domain second signal element, and

reordering the identifiers corresponding to at least one frequency to improve the coherence of the frequency-domain second signals to produce a set of frequency-domain third signals.

2. The method of claim 1, wherein the separating the set of frequency-domain first signals comprises a blind-source separation method.

3. The method of claim 2, wherein the blind-source separation method is applied to the frequency-domain first signal elements corresponding to individual frequencies.

4. The method of claim 1, wherein the coherence comprises a function that reflects the co-activity of signals at different frequencies.

5. The method of claim 1, wherein the improving the coherence of the frequency-domain second signals comprises optimizing the coherence of the frequency-domain second signals.

6. The method of claim 1, wherein the improving the coherence of the frequency-domain second signals comprises improving the coherence of a pair of adjacent frequency-domain second signal elements.

7. The method of claim 6, wherein the reordering the identifiers comprises reordering the identifiers corresponding to a plurality of frequencies, and wherein the reordering the identifiers further comprises sequentially reordering the identifiers from a low frequency of the plurality of frequencies to a high frequency of the plurality of frequencies.

8. The method of claim 6, reordering the identifiers comprises reordering the identifiers corresponding to a plurality of frequencies, and wherein the reordering the identifiers further comprises improving the coherence between non-overlapping pairs of frequency-domain second signal elements.

9. The method of claim 1, wherein the improving the coherence of the frequency-domain second signals comprises:

optimizing the coherence of frequency-domain second signal elements corresponding to a first subset of frequencies; and

optimizing the coherence of frequency-domain second signal elements corresponding to a second subset of frequencies,

wherein the second subset of frequencies is larger than the first subset of frequencies, and

wherein the second subset of frequencies comprises the first subset of frequencies.

10. The method of claim 1, further comprising transforming a set of time-domain input signals to produce the set of frequency-domain first signals.

11. The method of claim 9, wherein the time-domain input signals are acoustic signal mixtures.

12. The method of claim 10, wherein the acoustic signal mixture comprises a speech signal.

13. The method of claim 1, further comprising inverse transforming a frequency-domain third signal of the set of frequency-domain third signals to produce a desired signal.

14. A signal separation method, comprising

receiving a set of frequency-domain first signals,

wherein each frequency-domain first signals comprises frequency-domain first signal elements corresponding to a plurality of frequencies, and

wherein the set of frequency-domain first signals comprises a plurality of signal sources;

separating the set of frequency-domain first signals into a set of frequency-domain second signals using an Independent Vector Analysis (IVA) method comprising a learning rule; and

applying one or more constraints to the learning rule.

15. The method of claim 14, wherein the one or more constraints are derived from the set of frequency-domain second signals.

16. The method of claim 14, wherein the applying one or more constraints to the learning rule reduces the probability of converging upon a local minimum or maximum.

17. The method of claim 14, further comprising receiving a second set of frequency-domain first signals and separating the second set of frequency-domain first signals using the constrained signal separation method.

18. The method of claim 14, wherein the one or more constraints comprise geometric constrains.

19. The method of claim 18, wherein the geometric constraints comprise estimated direction of arrival information and wherein the direction of arrival information is estimated by computing output signal subbands from histograms and/or other statistical techniques using known sensor locations.

20. The method of claim 14, wherein the one or more constraints comprise information regarding spectral or energy characteristics of a signal.

21. The method of claim 20, wherein the signal is selected from frequency-domain first signal, a desired output signal, and an undesired output signal.

22. The method of claim 14, wherein the one or more constraints comprise non-derived constraints.

23. A system for separating signals, comprising:

a receiver that receives a set of time-domain input signals;

a transformer module that transforms the set of time-domain input signals to a set of frequency-domain first signals,

wherein each frequency-domain first signals comprises frequency-domain first signal elements corresponding to a plurality of frequencies, and

wherein the set of frequency-domain first signals comprises a plurality of signal sources;

a frequency-domain signal separator module that separates the set of frequency-domain first signals into a set of frequency-domain second signals,

wherein the set of frequency-domain second signals comprises a set of separated frequency-domain second signal elements corresponding to individual frequencies of the plurality of frequencies, and

wherein each frequency-domain second signal element is assigned an identifier which indicates which of the set of frequency-domain second signals includes the frequency-domain second signal element, and

a reordering module that reorders the identifiers corresponding to at least one frequency to improve the coherence of the frequency-domain second signals to produce a set of frequency-domain third signals.

24. The system of claim 23, wherein the frequency-domain signal separator module separates the set of frequency-domain first signals by using an Independent Component Analysis method or Independent Vector Analysis method.

25. The system of claim 23, wherein the reordering module comprises:

an initial parameter sub-module that determines the first frequency-domain second signal elements to be reordered by determining the corresponding frequency;

a coherence improving sub-module that improves the coherence of frequency-domain second signal elements corresponding to a predetermined frequency with frequency-domain second signal elements corresponding to a frequency adjacent to the pre-determined frequency; and

a frequency-advance sub-module that determines subsequent frequency-domain second signal elements to be reordered, wherein the frequency-advance sub-module determines subsequent frequency-domain second signal elements as those corresponding to a frequency adjacent to a frequency corresponding to the most-recently reordered frequency-domain second signal elements.

26. The system of claim 23, wherein the reordering module comprises:

an initial parameter sub-module that determines the first frequency-domain second signal elements to be reordered by determining the corresponding frequency;

a coherence improving sub-module that improves the coherence of frequency-domain second signal elements corresponding to a pre-determined frequency with frequency-domain second signal elements corresponding to a frequency adjacent to the pre-determined frequency; and

a frequency-advance sub-module that determines subsequent frequency-domain second signal elements to be reordered, wherein the frequency-advance sub-module determines subsequent frequency-domain second signal elements as those not previously analyzed by the coherence improving sub-module.

27. The system of claim 23, wherein the reordering module comprises:

an initial parameter sub-module that determines the initial state of a group of reordering elements as first frequency-domain second signal elements corresponding to at least one frequency;

a coherence improving sub-module that improves the coherence of frequency-domain second signal elements within the group of reordering elements; and

a frequency-advance sub-module that determines frequency-domain second signal elements to be added to the group of reordering elements.

28. The system of claim 23, further comprising an inverse transformer module that inversely transforms a frequency-domain third signal of the set of frequency-domain third signals to produce a desired signal.

29. The system of claim 27, wherein the system comprises a headset.

30. The system of claim 27, wherein the system comprises a telephone.

31. A system for separating signals, comprising:

a receiver that receives a set of time-domain input signals;

a transformer module that transforms the set of time-domain input signals to a set of frequency-domain first signals,

wherein each frequency-domain first signals comprises frequency-domain first signal elements corresponding to a plurality of frequencies, and

wherein the set of frequency-domain first signals comprises a plurality of signal sources;

a frequency-domain signal separator module that separates the set of frequency-domain first signals into a set of frequency-domain second signals using an Independent Vector Analysis (IVA) method comprising a learning rule; and

a constraint imposer module that imposes one or more constraints on the learning rule.

32. The system of claim 31, further comprising a constraint deriver module that derives the one or more constraints from the set of frequency-domain second signals.

33. The system of claim 31, wherein the one or more constraints comprise geometric constrains.

34. The system of claim 33, wherein the geometric constraints comprise estimated direction of arrival information.

35. The system of claim 34, wherein the direction of arrival information is estimated by computing output signal subbands from histograms and/or other statistical techniques using known sensor locations.

36. The system of claim 31, wherein the one or more constraints comprising non-derived constraints.

37. The system of claim 31, wherein the one or more constraints comprises information regarding spectral or energy characteristics of a signal in the frequency-domain signal separator module.

38. The system of claim 31, wherein the system comprises a headset.

39. The system of claim 31, wherein the system comprises a telephone.