High clarity speech separation using synchro extracting transform

In an era of ever improving communication technologies, many algorithms have been developed for recovering speech signals which provide more intelligible and listenable signals from their mixtures without any prior information about the signal being separated. Degenerate Unmixing Estimation Technique (DUET) is a Blind Source Separation (BSS) method which is highly suitable for underdetermined conditions wherein the number of sources exceeds number of mixtures. Estimation of mixing parameters which is the crucial part of DUET algorithm is built based on the idea of sparseness of speech signal in Time Frequency (TF) domain. Hence DUET is heavily dependent on the clarity of Time Frequency Representation (TFR) and any interference terms in TF plane will affect its performance adversely. Short Time Fourier Transform (STFT) is used to convert speech signals to TF domain in conventional DUET algorithm.STFT has its own limitations in providing sharpness to TFR due to its inherent characteristics which worsens with noise contamination. The paper presents a method of post-processing based on Synchro Squeezed Transform (SST) and Synchro Extracting Transform (SET) techniques to improve TF resolution of DUET method.The efficiency of these methods are evaluated both qualitatively and quantitatively by visual inspection,Renyi entropy of TFR and objective measures of speech signals.The results indicate how the sharpness of TFR provided by these transforms can improve the ability of signal reconstruction and robustness to noise which in turn improves the clarity of reconstructed signal.