Abstract: Our study consists in conceiving a compression of system of the speech signal by the vector quantization. The speech signal is analyzed by the method of Prediction Linear LPC which allows to converter this signal in acoustic vectors. Our system functions in two phases; the phase of training and the phase of coding. In the first phase, a dictionary of prototype vectors is generated by new robust algorithm. During the phase of coding, the speech signal undergoes the same acoustic treatment. Each acoustic vector is quantified and coded by the index of the prototype vector nearest to the vector of entry within the meaning of metric selected. Then, a binary code is attributed to each index. Thus, our system transforms the speech signal into a sequence of index and strongly contributes to its compression.

Prediction linear LPC. A dictionary of prototype. Speech signal.