High quality 0.6 kb/s speech coding algorithm
Junlin Li · Journal of Tsinghua University(Science and Technology) · 2003
A 0.6 kb/s high quality vocoder was developed to encode phonetic information at very low bit rates. The algorithm is based on a sinusoidally excited linear prediction model and uses multi frame joint vector quantification, multi band mixing excitation, sub band voicing strength parameter prediction, and adaptive spectral enhancement to obtain high quality synthetic speech with a low bit rate. Simulation results show that the synthesized speech is intelligible with reasonable naturalness. The diagnostic rhyme test score was 89.5% in the formal test. The vocoder is robust in a noisy environment and is still intelligible with a bit error rate of 10 -2 . The results suggest that the use of relative frame parameters and vector qualitification can greatly reduce the bit rate while maintaining clarity.