A 4.1 kb/s hybrid speech coder
Mohammad Reza Nakhai, Farokh Marvasti · 2003
In this paper, a hybrid speech coder based on a mixture of sinusoidal and CELP coding algorithms is introduced. The conventional harmonic sinusoidal coding algorithm is revised in terms of fundamental frequency estimation, unvoiced classification and decomposition of speech spectrum into two regions of harmonic and non-harmonic components. Sinusoidal and CELP coding algorithms are used to code the harmonic and non-harmonic components, respectively. Peaks of speech spectrum are, first, interpolated and, then, represented by linear predictive all-pole filter parameters. A harmonic tracking algorithm interpolates the sinusoidal parameters between adjacent frames to introduce an essential level of periodicity in the synthetic speech waveform. Pure unvoiced frames are only coded by CELP algorithm. Spectrograms of the original speech and its synthetic versions synthesised by the proposed hybrid speech coder operating at 4.1 kb/s and the Motorola GSM half-rate speech coder operating at 5.6 kb/s are compared in both noise-free and noisy environments.