Speech source localization sub-space algorithm research based on microphone array in near field

The speech source localization technology based on microphone array has been abroad applied to communication and speech signal processing.. such as teleconference, hands-free mobile telephone, hearing aids, etc. It has been proposed as a promising solution that can significantly increase the quality of speech Communication in the noisy background, and will replace conventional desktop and head microphones in near future. In array signal processing, a plane wave and white Gauss noise is assumed for the sake of simplicity (far-field and independence signal sources assumption). However, this far-field does not always hold in microphone array, resulting in distortion in the array output. The performance of DOA estimation will go bad under nonwhite Gauss noise circumstance. An accurate near-field signal model and a precise factor about relative amplitude attenuation are obtained from the diffusing principle of speech in this paper. The audio signal is referred to as wideband signals since the ratio of highest to lowest frequency component is quite large. On the other hand, most propagated RF signal are narrowband. since the ratio of the highest frequency to the lowest frequency is usually very Close to unity. Narrowband signals have a well-defined nominal wavelength, and wideband signals are no characteristic wavelength. To find the location of speech source, a novel modified 3D (three dimension) MUSIC method is proposed in this paper. The main step of this method is: a) Transform the time domain signal into frequency domain signal by STFT(Short Time Fourier-Transform); b) Divided the wideband signal into several sub-band signals according to the signal frequency, and each sub-band is narrowband signal with a characteristic wavelength; c) Chosen proper data length in order that each sub-band signal is stationary. d) Calculated the correlation matrix of each sub-band signal; e) Decomposed this correlation matrix, and the noise subspace and signal subspace can be gotten; f) Define the space spectrum matrix and search the matrix in three dimension to Find the location of speech source. g) Average these results by each sub-band with especial weight. and the estimation of speech source location is gotten. This method assumes that the noise spectrum would be zero mean complexes white Gaussian distributed in each element. The performance of localization is bad when the interference signal is present by this method. To modify the array topology and the correlation matrix of each sub-band signal, a novel sub-array algorithm based oil sub-space method is proposed in this paper. Simulation results show the high performance of the localization is gotten when the modified MUSIC algorithm is used in-planar uniform circle array and 3D uniform spherical surface array when the interference signal isn't present. When the interference exists, the subarray algorithm has more performance than the modified MUSIC algorithm.