Research Of Low Bit-Rate Sinusoidal Transform Coding

The frequency resource becomes more valuable along with high-speed development of communication technique. Compressing the bandwidth of speech signal in a digital speech communication system with speech coding technique, especially with low bit-rate speech coding technique, is still a hotpot which is concerned in the communication field now and in the future. In this thesis, the research aims at the low bit-rate algorithm of Sinusoidal Transform Coding (STC), and then a STC codec with a low bit-rate of 1.4kb/s is designed.Firstly, the researching background of speech coding and researching actuality are introduced. Various existing methods of speech coding and their applications in speech coding standards are also classified and summarized.Secondly, the sinusoidal model, which is the speech production model used in STC, and its simplified model, harmonic sinusoidal model, are depicted in this thesis. Then, the key techniques, including pitch detection, magnitude model presentation, phase model presentation, voice probability analysis, vector quantization and speech synthesis, are discussed.Therein, an improved harmonic production spectrum pitch detection algorithm is proposed for the low bit-rate STC. The computation complexity of the improved algorithm is greatly reduced while the high precision of original algorithm is preserved. For the magnitude model representation, a discrete all pole (DAP) model is introduced to estimate the magnitude parameters of the harmonic sinusoidal model. The theoretical analysis and simulation prove it is better than linear prediction magnitude envelope estimating, which is usually adopted in low bit-rate STC. With the phase model representation, the phase parameters are produced directly by phase prediction model depending on voice probability in the STC decoder without coding and transmitting the phase parameters, so the bit-rate can be reduced. To quantize and code the parameters needed transmission in STC coder, a perception weighting split vector quantizer...