论文部分内容阅读
我们在本文中考察了将分段声码器的比特率压缩百分之三十五到150b/s的若干种方法。在原始的声码器中,我们把矢量的随机抽样作为一组样板来进行矢量量化。在本文中,我们通过把随机量化器与采用分组量化算法进行语言分段量化的量化器进行对比,论证了随机量化器为近最佳量化器。分段声码器比特率能实现压缩,主要是由于使用了分段网络,即并不是使所有的分段样板都能跟踪上已知的分段样板。语言谱的连续性被用于确定样板的子集,它可以用于量化输入段。为了获得150b/s的低速率,我们同时还压缩了编码基音、增益和分段长度的比特率。在文章最后,我们介绍一下作为单一讲话人分段声码器以150b/s传输语言时的比特分配情况。
In this paper, we examine several ways to compress the bit rate of a segment vocoder between 35 and 150b / s. In the original vocoder, we random vector samples as a set of templates for vector quantization. In this paper, we demonstrate that a random quantizer is a near-best quantizer by comparing a random quantizer with a quantizer that uses packet quantization to segment the speech. The ability to compress the bit-rate of the vocoder is mainly due to the use of a fragmented network, ie not to allow all the fragmented samples to track the known fragmented samples. The continuity of language spectrum is used to determine a subset of the template that can be used to quantify the input segment. In order to get a low rate of 150b / s, we also compress the bit rates of encoding pitch, gain, and segment length. At the end of the article, we introduce the bit allocation when transmitting speech at 150b / s as a single speaker segment vocoder.