초록 ▼

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. A mixture model is used in which the observed data is categorized into two or more mutu

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. 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. 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.

대표청구항 ▼

1. A system for separating an independent source signal from a mixture of source signals, comprising:a plurality of sensors configured to receive said mixture of source signals;a processor configured to take samples of said mixture of source signals over time and store each sample as a data vector t

1. A system for separating an independent source signal from a mixture of source signals, comprising:a plurality of sensors configured to receive said mixture of source signals;a processor configured to take samples of said mixture of source signals over time and store each sample as a data vector to create a data set;a first function configured to assign each data vector within said data set to one class within a plurality of classes based on its similarity to other data vectors in said one class; andan independent component analysis module configured to perform an independent component analysis of the data vectors assigned to said one class to separate an independent source signal from other signals in said mixture of source signals. 2. The system of claim 1, wherein said plurality of sensors comprises a plurality of microphones and said mixture of source signals comprises a mixture of sounds. 3. The system of claim 1, wherein said system comprises an image processing system, and said mixture of source signals comprises a mixture of images. 4. The system of claim 1, wherein said system comprises a medical data processing system and said mixture of source signals comprises a mixture of medical data signals. 5. The system of claim 1, wherein said system comprises a speech compression system and said independent source signal comprises a voice signal. 6. The system of claim 1, wherein said system comprises a speech recognition system and said independent source signal comprises a voice signal. 7. The system of claim 1, wherein said first function is configured to generate class parameters for each data vector and for each class within said plurality of classes. 8. The system of claim 7, wherein said first function is configured to assign a data vector to said one class by comparing the class parameters of said data vector to the class parameters of said class. 9. The system of claim 8, wherein said first function is configured to create a new class if the class parameters of said data vector do not match the class parameter of any class in said plurality of classes. 10. A system for separating an individual feature of an image from a mixture of features in the image, comprising:a processor configured to take samples of said image, wherein said samples are stored to a memory as image data vectors;a first function configured to assign each image data vector to one class within a plurality of classes based on its similarity to other image data vectors in said one class; andan independent component analysis module configured to perform an independent component analysis of the image data vectors assigned to said one class to separate an individual image feature from other features in said image. 11. The system of claim 10, wherein said samples of said image comprises a set of image pixels. 12. The system of claim 10, wherein said first function is configured to generate class parameters for each image data vector and for each class within said plurality of classes. 13. The system of claim 12, wherein said first function is configured to assign an image data vector to said one class by comparing the class parameters of said image data vector to the class parameters of said class. 14. The system of claim 13, wherein said first function is configured to create a new class if the class parameters of said image data vector do not match the class parameter of any class in said plurality of classes. 15. The system of claim 10, comprising a memory configured to store predetermined class parameters for said plurality of classes. 16. The system of claim 15, wherein the first function is configured to assign an image data vector to said one class by comparing the class parameters of said image data vector to the predetermined class parameters. 17. The system of claim 10, wherein said image is a digital image and said individual feature is a bar code. 18. The system of claim 10, wherein said image comprises image data and text data and said in dividual feature comprises said text data. 19. The system of claim 10, wherein said image comprises image data and noise data, and said individual feature comprises said image data. 20. A method for separating an independent source signal from a mixture of source signals, comprising:receiving mixture of source signals from a plurality of sensors;taking samples of said mixture of source signals over time;storing said samples as a data set, wherein each of said samples comprises a data vector within said data set;assigning each data vector within said data set to one class within a plurality of classes based on its similarity to other data vectors in said one class; andperforming independent component analysis of the data vectors assigned to said one class to separate an independent source signal from other signals in said mixture of source signals. 21. The method of claim 20, wherein said plurality of sensors comprises a plurality of microphones and said mixture of source signals comprises a mixture of sounds. 22. The method of claim 20, wherein said mixture of source signals comprises a mixture of image signals. 23. The method of claim 20, wherein said mixture of source signals comprises a mixture of medical data signals. 24. The method of claim 20, wherein said independent source signal comprises a voice signal. 25. The method of claim 20, wherein said assigning each data vector to one class comprises generating class parameters for each data vector and for each class within said plurality of classes. 26. A memory comprising instructions that when run perform a method comprising:receiving mixture of source signals from a plurality of sensors;taking samples of said mixture of source signals over time;storing said samples as a data set, wherein each of said samples comprises a data vector within said data set;assigning each data vector within said data set to one class within a plurality of classes based on its similarity to other data vectors in said one class; andperforming independent component analysis of the data vectors assigned to said one class to separate an independent source signal from other signals in said mixture of source signals. 27. The memory of claim 26, wherein said plurality of sensors comprises a plurality of microphones and said mixture of source signals comprises a mixture of sounds. 28. The memory of claim 26, wherein said mixture of source signals comprises a mixture of image signals. 29. The memory of claim 26, wherein said mixture of source signals comprises a mixture of medical data signals. 30. The memory of claim 26, wherein said independent source signal comprises a voice signal. 31. The memory of claim 26, wherein said assigning each data vector to one class comprises generating class parameters for each data vector and for each class within said plurality of classes.