Separation of a mixture of acoustic sources into its components

• US 6,317,703 B1
• Filed: 10/17/1997
• Issued: 11/13/2001
• Est. Priority Date: 11/12/1996
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Linsker, Ralph
• Primary Examiner(s): Isen, Forester W.
• Assistant Examiner(s): PENDLETON, BRIAN T
• Application Number:
  US08/953,591
• Time in Patent Office: 1,488 Days
• Field of Search: 381/94.1, 381/94.3, 381/71.1-71.14, 702/190, 702/191, 367/119, 367/121
• US Class Current: 702/190
• CPC Class Codes: H04R 2201/403 Linear arrays of transducers
• CPC Class Codes: H04R 25/405 by combining a plurality of transducers
• CPC Class Codes: H04R 25/407 Circuits for combining signals of a plurality of transducers
• CPC Class Codes: H04R 3/005 for combining the signals of two or more microphones specially adapted for hearing aids H04R25/407

Abstract

A method and apparatus for processing a composite acoustic signal to reconstruct an acoustic signal that substantially matches a selected one of a plurality of sources. A plurality of microphones positioned at different spatial locations detect. variations in sound pressure level resulting from the activity of a plurality of acoustic sources at different locations. The outputs of the microphones are sampled and digitized, and the resulting digital waveform from each microphone is provided as an input to a corresponding filter bank. The outputs of the filter banks are input to a comparison unit. A comparison control unit generates “signature” information that characterizes each source with respect to the microphones. The comparison unit receives “signature” information of a selected source from the comparison control unit and provides an output to a synthesizer unit which produces a synthesized digital waveform for the selected source. Optionally, the synthesized digital waveform is input to a digital-to-analog (D/A) converter to generate an analog signal of the reconstructed source.

Claims

1. A signal processing method which reconstructs an acoustic signal that substantially matches a selected one of a plurality of sources comprising the steps of:

creating a time-frequency representation of a composite acoustic signal generated by said plurality of sources;

comparing selected regions of the time-frequency representation;

assigning a plurality of non-zero energy regions of the compared regions to a single source wherein, for at least a first of said non-zero energy regions that is assigned to a single source there are at least second and third non-zero energy regions that are not assigned to said single source, such that

(a) said second non-zero energy region shares the same frequency range as said first non-zero energy region; and

(b) said third non-zero energy region shares the same time range as said first non-zero energy region;

and

reconstructing the selected one of the plurality of acoustic sources from the set of assigned non-zero energy regions.

2. A signal processing method which reconstructs an acoustic signal that substantially matches a selected one of a plurality of sources comprising the steps of:

creating a time-frequency representation of a composite acoustic signal generated by said plurality of sources;

comparing selected regions of the time-frequency representation using pitch repeat information from the time-frequency representation;

assigning a plurality of non-zero energy regions of the compared regions to a single source; and

reconstructing the selected one of the plurality of acoustic sources from the set of assigned non-zero energy regions.

3. A signal processing method which reconstructs an acoustic signal that substantially matches a selected one of a plurality of sources comprising the steps of:

(a) detecting at each of a plurality of locations a composite acoustic signal;

(b) sampling and digitizing the detected composite acoustic signals to generate a plurality of digital waveforms;

(c) digitally filtering the digital waveforms to produce filter output values at each of a plurality of discrete times, a set of filter output values over a plurality of times constituting a filter output pattern, each filter output value of a filter output pattern being uniquely identified by an index of a filter that generated that filter output value and a time at which it was generated;

(d) generating control information including a set of comparison parameters;

(e) comparing output values having indexes and times that are specified by said control information by computing a function of these quantities and comparison parameters and determining whether or not the output values are a match;

(f) if a result of the comparison is a match, using the output values to compute a filter target value;

(g) repeating steps (e) and (f) a plurality of times to generate a set of filter target values which, taken as a whole, form a filter target pattern; and

(h) using the filter target values to produce a synthesized digital waveform that has the property that if the synthesized digital waveform were filtered, the resulting output pattern would be similar to the filter target pattern at those positions where the filter target pattern is defined.

4. The signal processing method recited in claim 3 wherein, in step (e), the indexes and times specified by the control information are the same index and time for each of the filter output values that are compared to one another.

5. The signal processing method recited in claim 3 further comprising the step of selecting digital filters used in the step of digitally filtering by choosing filters with a measured parameter of a degree of a sparse-representation property that is made large over a range of frequencies and for a relevant type of sounds.

6. The signal processing method recited in claim 5 wherein the measured parameter of the degree of the sparse-representation property of the filters is the kurtosis.

7. The signal processing method recited in claim 5 wherein the selected filters are substantially log-Gabor filters.

8. The signal processing method recited in claim 5 wherein the selected filters are substantially Gabor filters.

9. The signal processing method recited in claim 3 wherein the step of generating control information comprises the steps of:

selecting a value of gain factor and phase shift for each frequency for a source to be reconstructed;

tracking a gain factor/phase shift pair over time; and

outputting information based on the selected and tracked gain factor/phase shift pair a signature of the source to be reconstructed.

10. The signal processing method recited in claim 3 further comprising the step of converting the synthesized digital waveform to produce an analog synthesized acoustic signal.

18. The signal processing method recited in claim 3, wherein the number of said sources is greater than the number of said locations.

11. An acoustic signal processing apparatus for reconstructing an acoustic signal that substantially matches a selected one of a plurality of sources comprising:

a plurality of microphones positioned at different spatial locations detecting variations in sound pressure level resulting from the activity of a plurality of acoustic sources at different locations;

a plurality of sampling and digitizing units, one for each said microphone, sampling and digitizing detected variations in sound pressure levels at each said microphone to produce digital waveforms from each microphone;

a plurality of filter banks each respectively receiving a digital waveform from each microphone and producing filter output values at each of a plurality of discrete times, a set of filter output values over a plurality of times constituting a filter output pattern, each filter output value of a filter output pattern being uniquely identified by an index of a filter that generated that filter output value and a time at which it was generated;

a comparison unit receiving outputs from the plurality of filter banks;

a comparison control unit generating signature information that characterizes at least one source with respect to the microphones and supplying the signature information of a selected source to the comparison unit, said comparison unit comparing output values having indexes and times that are specified by said signature information by computing a function of these quantities and comparison parameters and determining whether or not the output values are a match, and if a result of the comparison is a match, using the output values to compute a filter target value, thereby generating a set of filter target values which, taken as a whole, form a filter target pattern; and

a synthesizer unit receiving the filter target pattern from the comparison unit and producing a synthesized digital waveform for the selected source.

12. The acoustic signal processing apparatus recited in claim 11 wherein the filter banks comprise digital filters having a measured parameter of a degree of sparse-representation that is made large over a range of frequencies and for a relevant type of sounds.

13. The acoustic signal processing apparatus recited in claim 12 wherein the measured parameter of the degree of sparse-representation of the digital filters is the kurtosis.

14. The acoustic signal processing apparatus recited in claim 12 wherein the digital filters are substantially log-Gabor filters.

15. The acoustic signal processing apparatus recited in claim 12 wherein the digital filters are substantially Gabor filters.

16. The signal acoustic signal processing apparatus recited in claim 11 wherein the comparison control unit comprises:

means for selecting a value of gain factor and phase shift for each frequency for a source to be reconstructed;

means for tracking a gain factor/phase shift pair over time; and

means for outputting information based on the selected and tracked gain factor/phase shift pair as the “

signature”

of the source to be reconstructed.

17. The acoustic signal processing apparatus recited in claim 11 further comprising a digital-to-analog (D/A) converter connected to receive the synthesized digital waveform from the digital synthesizer unit to generate an analog signal of the reconstructed source.

19. The acoustic signal processing apparatus recited in claim 11, wherein the number of said sources is greater than the number of said microphones.