摘要： 在混合激励线性预测(MELP)算法的基础上，设计一种1 120 b/s MELP甚低速率语音编码算法。该算法通过增加帧长、动态比特分配和多帧联合矢量量化、及参数内插等方法降低语音的编码速率，并已在TMS320VC5416DSP芯片上实时实现。采用美国GL公司的语音质量评估系统VQT，对编解码的实时语音质量进行评估，语音质量感知评价高于3分。实验结果表明，该算法能够满足实际通信要求。

关键词: 甚低速率语音编码, 动态比特分配, 多帧联合矢量量化, 参数内插

Abstract: Based on the MELP algorithm, an improved 1 120 b/s MELP very coding algorithm is proposed and implemented. Increase of framesize, dynamic bit allocation, multi-frame joint vector quantization and parameter interpolation are used to reduce the bit rate. The speech coding algorithm is implemented on DSP chip TMS320VC5416 in real time. Tested by GL Voice Quality Tester(VQT), the Perceptual Evaluation of Speech Quality(PESQ) scores of the speech coder is more than 3 points. Experimental result shows that this speech coding algorithm can meet actual communication requirements.

Key words: very low bit rate speech coding, dynamic bit allocation, multi-frame joint vector quantization, parameter interpolation

中图分类号: 

• TP391