一种选择预测矢量量化的迭代优化设计方法

Iterative approach for design of switched-predictive vector quantizers

为了提高选择预测矢量量化的性能,提出了一种迭代优化设计方法。利用常规的前后帧相关性划分训练数据,采用不同的训练数据进行各个预测系数计算和码本设计。采用设计的预测系数和码本对训练数据进行量化,根据量化误差大小调整训练数据划分。通过训练数据划分和码本设计迭代,优化选择预测矢量量化码本设计。该方法改进了训练数据固定划分的缺点,可以有效提高选择预测矢量量化设计性能。以语音线谱对参数为实验数据进行实验,实验结果表明,该方法能减小参数量化失真,改善信号压缩质量。

An iterative design approach of switched-predictive vector quantizers was proposed. Firstly, training vectors were classified according to the correlation between consecutive vectors. Predictive vector quantizers were designed for each class. Then the training vectors were classified again according to the quantization distortion with the switched-predictive vector quantizer. Designing switched-predictive vector quantizer and classifying the training vectors were alternated. The switching mechanism can improve the performance of the design. With linear spectrum pairs of speech as experimental data, the simulation re- sults show that the iterative can lower the quantization distortion and improve the quality of signal com- pression.