Research On Linear Predictive Speech Coding Based On Sinusoidal Model At Low-Bit-Rates

With the development of the Telecommunication network, high quality speech coding at low bit rates has become research focus. Some new methods are presented in this paper to improve the quality of Harmonic Excited Linear Predictive (HELP) coding which Telecommunication and Signal Processing Laboratory in Beijing Polythenic University develop.First, after deeply investigating HELP model, a harmonic related voicing detection algorithm based on MSE criterion is developed, with the knowledge that voicing algorithm can be showed by degree of harmonic relation. Second, efficient pitch detection is developed in frequency domain, which find fractional pitch. Finally, harmonic phase reconstruction that exploits minimum-phase model is proposed. Computer simulation and subject listening test show that the algorithm of this paper can efficiently improve the quality of reconstructed speech of 2.4kb/s HELP speech coding.