论文部分内容阅读
本文讨论了人工神经网络技术在语音压缩编码方面的应用,提出了一种用Kohonen网络实现语音多脉冲激励分析模型的矢量量化方法。该方法将参数分析和量化编码熔为一体,和传统的先分析、后量化方法相比较,具有许多优良的特性,如全并行处理、过程简化等。本文针对语音多脉冲激励模型,提出了量化网络的结构和学习规则,并将此方法和传统方法进行了比较。最后对网络的压缩性能进行了计算机模拟,结果表明应用人工神经网络进行语音信源的压缩是切实可行的。
This paper discusses the application of artificial neural network technology in speech compression coding, and proposes a vector quantization method for speech multi-pulse excitation analysis model using Kohonen network. Compared with the traditional methods of pre-analysis and post-quantification, the proposed method integrates parameter analysis with quantitative coding, and has many excellent features, such as full parallel processing, simplified process and so on. In this paper, aiming at the multi-pulse speech excitation model, the structure and learning rules of the quantified network are proposed, and the method is compared with the traditional method. Finally, the network compression performance is simulated by computer. The result shows that it is practicable to apply artificial neural network to compress the speech source.