CELP with priority to critical segments

At low bit rates (around and below 4 kbit/s), the use of a rigid coding configuration in the CELP (Code Excited Linear Prediction) speech coding structure leads to a bad reconstruction of the voiced onsets. In this paper we describe a 4 kbit/s CELP algorithm that gives priority to these critical speech segments. Our scheme employs different combinations of codebooks and bit allocations for different speech classes. Mean opinion scores (MOS) reveal that at 4 kbit/s the speech quality obtained with our coding algorithm is significantly better than that achieved with a CELP codec that does not use different coding strategies for different classes of speech.