FIELD: electric communications.
SUBSTANCE: method uses acoustic-phonetic classification of processed speech signal cadres as four non-crossing classes - absence of speech, vocalized speech, non-vocalized speech and transfer to vocalized speech. This classification is made concurrently with identification of filter of short-time linear prediction, and on basis of results information resources are adaptively distributed among encoded parameters. Classification decision is included into code combination structure for transfer along communication channel and is used to determine modes of vector quantizer and dequantizer, trained for each class of voice cadres.
EFFECT: higher quality.
2 cl, 8 dwg
