An adaptive algorithm for compression of speech signals

(new adaptive codec, which allows to significantly reduce the flow of data transmitted with a slight deterioration in the quality of the reconstructed speech at the receiving end)

In most modern applications that use voice transmission over the Internet data applied parametric coding algorithms (CELP, ACELP, LD-CELP), which can significantly increase the speed of transmission of speech information due to deterioration. One of the main problems of these methods of coding is to choose the most effective method of compression transition, voiced and unvoiced segments. There are many methods of solving this problem, but so far none of them provides a high quality MOS (Mean Opinion Score), due to incorrect encoding transient segments.

The most common orthogonal transformation used in various speech compression algorithms and image signals, a wavelet transform. Properties of the time-frequency localization and well-designed algorithms for rapid conversion makes a wide use of the wavelet transform in the analysis of non-stationary signals..

A new algorithm, which is based on the work of the classifier, allowing to allocate transitional (explosive, silence, unvoiced, affricate), voiced and unvoiced segments, followed by step selection of the most efficient orthogonal transformation, which can significantly reduce the flow of data transmitted over a digital communications channel with a slight deterioration of quality of the reconstructed speech at the receiver.