| Summary: | 碩士 === 大同工學院 === 電機工程研究所 === 87 === Amongst various speech coding schemes, the code-excited linear prediction (CELP) and its variants are most widely used in telecommunication applications. To meet different applications, the CELP-based variants were developed in tradeoff of bit rate, coding delay, and speech quality. In 1991, the International Telecommunication Union (ITU) has standardized the 16 kb/s low delay CELP (LD-CELP) as the recommendation G.728. However, to meet the more and more demanding commercial requirements, a low delay speech coding algorithm at 8 kb/s is mostly desired. Though in 1996, the 8 kb/s conjugate structure algebraic CELP (CS-ACELP) was accepted by the ITU as the recommendation G.729, it is still a medium delay system with a frame length of 10 ms.
In this thesis, we proposed a new 8 kb/s LD-ACELP speech coder with a frame length of only 4 ms. The proposed system utilizes a 10th order backward LPC predictor. The excitation signal is generated by two excitation codebooks, which include an adaptive codebook for pich predictor and a fixed codebook in algebraic form.
The proposed system utilizes the backward LPC predictor to save bit rate, since no LPC coefficients are necessary to be transmitted. The use of algebraic codebook effectively reduces the computational complexity. As to the quality of synthesized speech, our experimental results showed that the proposed system is significantly better than the 4.8 kb/s CELP and very close to the 16 kb/s LD-CELP.
|
|---|
