A NOVEL SPEECH ENHANCEMENT TECHNIQUE

R.Dhivya, Judith Justin

Abstract: This enhancement technique is a novel one and is based on the combination of Wavelet thresholding and Spectral Subtraction. Five wavelet filters are compared and the best filter is selected based on their performance of Signal to Noise Ratio. The selected filter is applied to the detail coefficients for thresholding. Approximation coefficient is applied to spectral subtraction filter. The reconstructed signal is evaluated using the metrics such as SNR (Signal to Noise Ratio), Correlation coefficient and PESQ (Perceptual Evaluation of Speech Quality). Real time data is recorded from Alaryngeal speakers and real world noise from Noizeus corpus is used for the study

Keywords: Wavelet Thresholding, Signal to Noise Ratio, Correlation coefficient, Perceptual Evaluation of Speech Quality

DOI: https://doi.org/10.15623/ijret.2014.0319019