Speech compression techniques based on the traditional psychoacoustic model have been proposed by many researchers. We propose the Discrete Wavelet Transform (DWT) supported by the same psychoacoustic model for speech compression. This paper presents a traditional psychoacoustic model for processing equal partitions of the total bandwidth spectrum of audio signal frequencies in order to reduce redundancy by filtering out the tones and noise maskers in the speech signal. Here, uniform filter banks are used for efficient computation, for selection of appropriate threshold levels, and for better compression of Discrete Wavelet Transform coefficients. A Daubechies wavelet filter bank is nonlinear and asymmetric. It is equivalent to a cochlear filter in the human hearing system. The similarity between the Daubechies filter bank and our hearing system was the basis for developing a novel speech coder. Better performance in terms of the compression factor (CF) and the signal-to-noise ratio (SNR) resulted, as compared to the earlier methods.

The paper presents integration of Discrete Wavelet Cosine Transform technique and Bacterial Foraging Algorithm (BFO) for the development and optimization of speech coder. It is depicted how by filtering the limited number of high energy components of transformed coefficients with parallel programming can maintain the speech signal quality in coding over wide range of bit rates. The performance of existing and proposed speech coding techniqueattributes such as compression ratio, coding delay, computational complexity and quality of reconstructed speech is examined for multiple bit rates and compared with other existing speech coding techniques in Matlab environment. The result showsimprovement in performancewith respect to all attributes at the cost of increase in complexity.

No abstract

No abstract