Unsupervised adaptation and classification of multiple classes and sources in blind signal separation

• US 6,424,960 B1
• Filed: 10/14/1999
• Issued: 07/23/2002
• Est. Priority Date: 10/14/1999
• Status: Expired due to Fees

• Current Assignee: Salk Institute
• Original Assignee: Salk Institute for Biological Studies
• Inventors: Lee, Te-Won, Sejnowski, Terrence J., Lewicki, Michael S.
• Primary Examiner(s): STARKS, WILBERT L
• Application Number:
  US09/418,099
• Time in Patent Office: 1,013 Days
• Field of Search: 706/20, 342/13, 375/262, 379/386, 600/310, 600/515
• US Class Current: 706/20
• CPC Class Codes: G06F 18/2134 based on separation criteria, e.g. independent component analysis
• CPC Class Codes: G06V 30/242 Division of the character sequences into groups prior to recognition; Selection of dictionaries
• CPC Class Codes: G10L 21/0272 Voice signal separating

Abstract

A computer-implemented method and apparatus that adapts class parameters, classifies data and separates sources configured in one of multiple classes whose parameters (i.e. characteristics) are initially unknown. The data set may be generated in a dynamic environment where the sources provide signals that are mixed, and the mixing parameters change without notice and in an unknown manner. A mixture model is used in which the observed data is categorized into two or more mutually exclusive classes. The class parameters for each of the classes are adapted to a data set in an adaptation algorithm in which class parameters including mixing matrices and bias vectors are adapted. Each data vector is assigned to one of the learned mutually exclusive classes. In some embodiments the class parameters may have been previously learned, and the system is used to classify the data and if desired to separate the sources. The adaptation and classification algorithms can be utilized in a wide variety of applications such as speech processing, image processing, medical data processing, satellite data processing, antenna array reception, and information retrieval systems. The adaptation algorithm described is implemented with an extended infomax ICA algorithm, which provides a way to separate sources that have a non-Gaussian (e.g., platykurtic or leptokurtic) structure.

Claims

1. A computerized method of separating an independent source signal from a mixture of source signals, comprising:

receiving said mixture of source signals into a plurality of data inputs of a computer,

reading a sample of said mixture of source signals from said data inputs;

comparing said sample to previously received samples of said source signals;

classifying said sample based on its similarity to other samples from said mixture of source signals; and

performing independent component analysis on said classified sample to separate an independent source signal in said sample from other signals in the class of signals.

2. The method of claim 1, wherein said mixture of source signals comprises a mixture of sounds.

3. The method of claim 2, wherein said mixture of sounds comprises a mixture of voices.

4. The method of claim 2, wherein said independent source signal comprises a single voice.

5. The method of claim 2, wherein said data inputs comprise microphones.

6. The method of claim 1, wherein said mixture of source signals comprises an EEG or an EKG.

7. The method of claim 1, wherein reading a sample of said mixture comprises reading a sample over a predetermined period of time.

8. The method of claim 1, wherein classifying said sample comprises:

determining data parameters that characterize the source in said other samples;

obtaining a first set of data parameters that characterize the source signals in said sample; and

assigning said sample to a first class based on the similarity of said first set of data parameters to the data pars for other samples in said first class.

9. A system for separating an independent source signal from a mixture of source signals, comprising:

a plurality of data inputs configured to receive a mixture of source signals into said system;

a memory in communication with the data inputs and configured to store samples of said mixture of source signals;

a first function that compares said samples to previously received samples of said source signals;

a second function that classifies each of said samples based on its similarity to other samples; and

an independent component analysis module that performs an independent component analysis of the source signals in each of said classes to separate an independent source signal in said sample from other source signals in said mixture of source signals.

10. The system of claim 9, wherein said mixture of source signals comprises a mixture of sounds.

11. The system of claim 10, wherein said mixture of sounds comprises a mixture of voices.

12. The system of claim 11, wherein said independent source signal comprises a single voice.

13. The system of claim 10, wherein said data inputs comprise microphones.

14. The system of claim 9, wherein said mixture of source signals comprises an EEG or an EKG.

15. The system of claim 9, wherein the samples of said mixture of source signals comprises samples taken over a predetermined period of time.

16. A system for separating a single voice signal from a mixture of voice signals, comprising:

a plurality of microphones configured to receive a mixture of voice signals into the system;

a memory in communication with the microphones and configured to store a sample of said voices;

a first function that compares said sample to previously received samples of said voice signals;

a second function that assigns said sample to a class of voice signals, wherein each sample in the class has similar data attributes;

an independent component analysis module that performs an independent component analysis of the voice signals in said class and separates a single voice signal from other voice signals in said class.

17. The system of claim 16, comprising a speaker that outputs said single voice signal that it separated from other voice signals in said class.

18. The system of claim 16, wherein the sample of said mixture of voice signals comprises a sample taken over a predetermined period of time.

19. The system of claim 16, wherein said system is a personal computer system.