实时低速率语音压缩编码的算法研究

针对数字化通信网及多媒体应用中低速率数字语音编码问题，以多带激励（ＭＢＥ）声码器为模型，采用了一些新算法去降低编码速率和改善音质。利用动态规划算法对基音周期进行平滑，去除了声码器中常有的音调噪声。利用ＬＰＣ全极点模型谱逼近ＭＢＥ谱包络，并采用共振峰增强技术来补偿模型误差，有效地降低了编码速率。为了能够实时实现这个编码系统，采用了分裂矢量量化，多级矢量量化和前向多层人工神经网络等技术进行优化和改善，使之在２．４ｋｂｉｔ／ｓ，１．２ｋｂｉｔ／ｓ及８００ｂｉｔ／ｓ等速率上实时实现了较高质量的语音压缩编码。

Research on real time low bit rate speech coding

Low bit rate speech coding is one of the most important and fundamental part of modern digital communication network and multimedia techniques. This paper presents a high quality variable rate speech coding at 2.4 kbit/s, 1.2 kbit/s and 800 bit/s based on the algorithm of multi band excitation (MBE). The techniques of smoothing pitch period using dynamic programming algorithm, LPC all pole model to approximate the MBE spectral shape, formant enhancement, split vector quantizer, multi stage vector quantizer, multilayer perception of neural networks, etc. are adopted to improve this vocoder. It is implemented in realtime by using a single ADSP2181 fixed point digital signal processor. The speech quality of the real time systems is rather satisfactory.