LD-CELP语音编码中一种新的码书设计方法

G．728语音压缩标准的算法延迟只有0．625毫秒，对于绝大多数应用来说是非常令人满意的。但是，其占用的带宽似乎稍微高了一点。为了能够降低G．728算法使用的码率，人们尝试了很多的方法，比如只采用前32个波形码字，采用奇数号的码字，等等。根据前人的研究，该文提出了一种新的12．8kbit／s编码器。该编码器利用人与人之间对码字使用的统计差异，自动地为每一个人生成一个独特的码书。这种编码器的计算复杂度与其它同码率编码器相当，内存使用有少量的增加，而音质则有比较明显的提高。

A New Codebook Design Method in LD-CELP Speech Coding Algorithm

The algorithmic delay of recommendation G.728 is low to 0.625ms, perfect for many applications. Yet its band width requirement seems a little high. In order to lower the bit rate, many methods have been tried, like using the first 32 waveform codevectors or the odd numbered codevectors. Based on the work done by others, we have devised a new codec which will work at 12.8kit/s with high synthesis speech quality. This codec takes advantage of the statistics difference among people and automatically creates a most-frequently-used codebook for each person. The algorithm complexity is similar to other codecs using the same bit rate, and it needs a little more RAM while the speech quality is improved notably.