The Research And Realization On Blind Separation Of Audio Mixtures

Blind Source Separation has been one of the new booming technologies over the past decades. It has made a big stride both in theories and practices and thus becomes one of the hot issues that attract worldwide attention. It is widely used in the domain of wireless telecommunications and image buildup and so on, in which many orientations and real effects have been manifested.This thesis mainly focuses on the research of BSS' application in audio separation. Firstly, the status quo and development level of BSS is analyzed, including the introduction of several theoretically ripe blind source separation algorithm. Secondly, the thesis gives a detailed analysis of the audio mixtures separation, including the separabi -lity of audio mixtures, different network structures of audio mixtures separation, analyses for audio's frequency ranges and the essence of frequency blind source separation. Also, in this section, FastICA is used to separate the audio of different frequency ranges The tests and analyses are included. Thirdly, introduction of the Self-adaptive filtered theory, the LMS algorithm and its convergence is analyzed; the tests and analyses are included too. In the last chapter, a brief introduction of Genetic Algorithm is given and the thesis proposes an optimized analysis model for the audio separation. The analysis model avoids the shortcomings of the traditional analysis which do not have a good as-tringency and the astringency is influenced by step-length.