论文部分内容阅读
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。
Aiming at the problem of low rate digital speech coding in digital communication network and multimedia applications, MBE vocoder is adopted as a model, and some new algorithms are adopted to reduce the coding rate and improve the sound quality. The dynamic programming algorithm is used to smooth the pitch period and remove the common pitch noise in vocoder. The LPC all-pole model is used to approach the MBE spectral envelope, and the formant enhancement technique is used to compensate the model error, which effectively reduces the encoding rate. In order to realize this coding system in real time, the techniques of split vector quantization, multi-level vector quantization and forward multi-layer artificial neural network are optimized and improved to make it suitable for 2.4kbit / s, 1.2kbit / s and 800bit / s real-time rate of higher quality voice compression coding.