Joint Noise Suppression and Dereverberation of separating speech signals by using prediction and separation matrix

Separation of speech signal is an difficult task for researchers for separation of speech and noise from the mixing signals in the field of speech processing. The mixing and unmixing estimation technique is used commonly in the blind source separation (BSS) algorithms; however there is many limitations in separating of speech and noises such as echoes and dereverberation from the speech signal. The removal of noises are very much significant of all related testing and training done in auditory system. In this paper an newly invented method has been introduced by joint noise suppression and dereverberation by using the separation and prediction matrix which acts as an processing element. Due to the analysis of prediction matrix and separation matrix the results show that this joint noise suppression and dereverberation provides good Signal to Interference Ratio (SIR) and Direct to Reverberation Ratio (DRR) as 3 dB and 1.8 dB respectively. By using the proposed method the recognition accuracy improves to 92%. The results shown that the new method achieving better accuracy, separation, noise separation and enhanced accuracy in separated speech signals.