论文部分内容阅读
混合激励线性预测(MELP)算法中,每帧5维的子带清浊音参数对于提高合成语音的自然度有着重要作用,但其每帧5 bit的编码效率给语音的极低速编码带来了困难。文章将MELP的3帧联合构成一个超级帧,对15维的子带清浊音参数进行矢量量化。通过清浊音信息的统计,并利用失真测度d进行码本的优化设计,实现了每个超级帧用3 bit对15维矢量的高效量化。仿真结果表明,文中算法对子带清浊音参数编码后,合成语音仍然保持了良好的可懂度和自然度,可有效应用于600 bps以下极低速语音编码算法中。
In the hybrid excitation linear prediction (MELP) algorithm, the five-dimensional sub-band unvoiced and voiced parameters per frame plays an important role in improving the naturalness of synthesized speech. However, the coding efficiency of 5 bits per frame makes it difficult to encode very low speed speech . In this paper, three frames of MELP are combined to form a super frame, and the 15-dimensional sub-band unvoiced and voiced parameters are vector-quantized. Through the statistics of the voiced and unvoiced information, and using the distortion measure d to optimize the codebook, the high-efficiency quantization of 15-dimensional vectors for each superframe is realized. The simulation results show that the algorithm still has a good degree of intelligibility and naturalness after encoding the sub-band unvoiced and voiced parameters, and can be effectively used in very low speed speech coding algorithms below 600 bps.