A two stage speech enhancement method for further improvement of speech quality by extracting signal from residual

A two stage speech enhancement technique is presented in this paper. We show that overall denoising efficiency can be further improved by incorporating a residual processing technique with a conventional speech denoising method. The reminiscent speech components extracted from the residual using a harmonic model in the voiced frames and a simple yet effective time-domain thresholding technique in the unvoiced frames are added to the first stage denoising filter output for further enhancement. The experimental results demonstrate significant improvement in speech quality indices over the single stage speech enhancement methods for a wide range of signal-to-noise ratios (SNRs).