The project consists in a voice coder (vocoder), splitted in two parts: the encoder and respectively the decoder. It is based on the LPC principle that, basically, is related to frame-based AR modelling of the speech (for frames of typically 20 ms, for which the speech signal is approximate stationary).

GSM like vocoder has some original improvements at pitch detection and at quantization (bit packing) part. Starting from a bitrate of 3400 bits/s, the quantized version is behaving like the non-quantized one.

Below there are some samples of voices (original and coded) that should give an insight about the possibility of speech coding by using this coder: