Continuous Speech Coding Using Coiflets Wavelet

Wavelet is a good choice for signal compression. Wavelets transform can represent signals more effectively than its predecessor transforms since it retain the time and frequency aspect of the signal. Main challenge in wavelets based speech coding is to choose the best wavelet for decomposition of the speech signal and optimum level of decomposition. Aim of this paper is to analyze the performance of coiflets wavelet in continuous Malayalam speech coding. This is the first attempt carried out among the Indian languages. In this experiment we used coiflets wavelet to decompose the signal and applied Birge-Massart strategy based thresholding to make insignificant wavelets coefficient to zero and finally lossless encoding algorithm applied to encode the remaining coefficients. Performance of this process evaluates in terms of SNR, PSNR, NRMSE, RSE, MOS and Compression ratio. For this experiment we used continuous spoken sentence from Malayalam, one of south Indian language. We could successfully compress and reconstructed continuos Malayalam spoken sentence with perfect audibility.