论文部分内容阅读
为在极低速率下实现高质量的语音编码,提出了一种新的有效的线谱对(LSP)参数量化算法——P-RS-MSMQ算法。此算法以多帧联合矩阵量化作为基本框架,引入了基于超级帧模式的均值去除和帧间预测策略、矩阵分裂和子矩阵多级量化策略;同时提出了基于语音帧短时谱能量的帧内加权和基于超级帧中各子帧重要性的帧间加权策略等。实验表明:此算法能够在700b/s的速率下获得接近透明量化的性能;即使在300~400b/s的极低速率下也具有较高质量的量化效果。因此该算法的实现对极低速率语音编码算法的研究具有重要的意义。
In order to achieve high-quality speech coding at very low rate, a new effective P-RS-MSMQ algorithm for spectral parameter estimation of line spectrum pair (LSP) is proposed. This algorithm takes the multi-frame joint matrix quantization as the basic framework, and introduces the mean removal and inter prediction methods based on the super-frame mode, the matrix splitting and the sub-matrix multi-level quantization strategy. At the same time, the intra-frame weighting based on the short- And an inter-frame weighting strategy based on the importance of each sub-frame in the super-frame. Experiments show that this algorithm can achieve near-transparent quantification performance at 700b / s and even higher quality quantization even at an extremely low rate of 300-400b / s. Therefore, the implementation of this algorithm is of great significance to the research of very low rate speech coding algorithms.