Single Channel Speech Noise Reduction Algorithm Based On Independent Component Analysis

Independent Component Analysis (for short ICA) is a new signal analysis method, which is derived from the developing Blind Sources Separation technology of multi-channel recently. ICA is a signal processing method on the condition that there only are observing signal data and unknown mixed method of the signals extracts independent source signals. The characteristic of the ICA is that the multi-channel's observed signals are decomposed to some mutual independent component according to statistical independent fundamental and optimization algorithm, consequently helps us to analyze signals and enhance signals'SNR. Now there are more and more concerns on using ICA algorithm to extract blind source signals. Mostly because the analysis method has potential advantage on audio recognition, communication, biomedicine signal processing and image processing. The dissertation mainly discuss processing low SNA speech signal base on ICA algorithm to enhance and reduce signal noise. Noisy speech can be regarded as some kind of mixture of noise and speech. Thereby noise and speech can be separated through ICA algorithm. In this way speech can be enhanced. ICA requires the number of observations to be no less than that of the independent sources. So it is impossible to apply ICA to a single signal directly. The paper presents a smart technique to generate two signals from one single signal, which can be processed by using ICA. Simulation results show that the ICA-based method can improve SNA in high noise condition and keep good intelligibility. Above all in chapter one, the author introduces independent component analysis algorithm's conception and application domain relates to the blind source signal processing, so we can basically understand ICA. In chapter 2, the author introduces independent component analysis...