A variable-rate natural-quality parametric speech coder

A variable bit rate speech coder operating below 2 kb/s is presented which combines phonetic classification and frequency domain modeling. Each input frame is identified as one of the following classes: mixed voiced, fully voiced, unvoiced, noise, and silence. Based on the class parameter, a suitable parametric coding scheme (spectral analysis, modeling, quantization, and synthesis) is selected. The coder thereby adaptively matches a suitable coding scheme to the character of the input speech leading to variable rates ranging from 0.15 kb/s (for silence) to 2.6 kb/s (for mixed voiced). For typical conversational speech, the coder operates at an average rate of 1.4 kb/s while delivering speech quality that is subjectively preferred over Federal Standard 1016 CELP at 4.8 kb/s by a majority of listeners.<
>