Enhanced Binaural Source Separation Using Wavelet Denoising Technique

This paper introduces a new enhanced binaural source separation procedure, which is the combination of a model-based source separator system, wavelet denoising technique, andminimum mean-square-error (MMSE) speech enhancement method. In the separation step, the parameters of interauralphase difference and interaural level difference are modeled as Gaussian mixture models (GMMs) for each source. By evaluatingthe model at each time-frequency (T-F) point, the units with high probabilities are assigned to the source. To improve the performance of the source separation system, we have proposed to use the wavelet denoising technique followed by the MMSE speech enhancement method as a two-step post-processing stage.Conducting different experimental simulations shows the superior performance of the proposed enhancement method in the context of model-based source separation systems

کلیدواژه ها:

binaural Source Separation ، interaural phase difference ، interaural level difference ، wavelet denoising technique ، MMSE speech enhancement