Enhancement Artificial Neural Networks for Low-Bit Rate Speech Compression System

Abstract

An Artificial Neural Networks (ANNs) is the intelligent system which has been recently exploited in linear and non-linear system such as image and speech processing. In this work, there are two types of neural networks, namely Kohonen Self Organizing Feature Maps (KSOFM) and Probabilistic Neural Networks (PNNs), which are investigated to use in CELP speech coding system. KSOFM is used to classify the repetitiveness of speech signal and create the optimal codebook and PNNs is applied to predict the codebook index by using the knowledge of training system. The results show that the neural index prediction can reduce the number of bit rate approximately 25% while maintains the quality of the synthesized speech as similar as the original speech.