摘要 : In the paper, an efficient speech-coding system based on sinusoidal model is implemented. Sinusoidal amplitudes and phases are represented by sampling the short time spectrum of an all-pole linear predictive coding (LPC) filter. A... 展开 In the paper, an efficient speech-coding system based on sinusoidal model is implemented. Sinusoidal amplitudes and phases are represented by sampling the short time spectrum of an all-pole linear predictive coding (LPC) filter. A gain adjustment module followed by a linear phase alignment module are then introduced to minimize a perceptually objective mean squared error criterion. The LPC filter is represented by line spectral frequencies (LSF). The derivation and encoding of system parameters is based on a large Arabic speech database. Quantization of LSF is based on the split vector quantization algorithm. Codebooks generated from different split schemes are compared then combined in a more efficient codebook. System modules are integrated into a 1300 bps sinusoidal encoder/decoder. Objective and subjective tests are used to evaluate the system performance. Computer simulation experiments show that the decoded speech output is of reasonable quality, intelligibility, and naturalness. 收起