Speech compression using ARMA model and wavelet transform

In LPC (linear predictive coding) vocoding, because of the insufficiency of the all-pole model, reconstructed speech is often crude unless much information about the residual error is also transmitted. To improve the quality, ARMA parameters of each bandpass filter output of a filter bank are extracted and transmitted in this paper. A general idea of the wavelet transform is utilized for design of the filter bank. Both the ARMA model and the wavelet transform are well known. But a combination of these technique has not been exploited in speech coding. This paper may lead to the new way for the very low-rate speech coding