Research And Implementation On Distributed Speaker Recognition System

Speech Recognition is the best way of the personal- machine communication. By the development of computer technology, it has been implemented into product in many areas such as control, communication and consumption. Not only is speech recognition a theoretic problem, but also an engineering problem. It integrates theoretic achievements of many disciplines, for example, acoustics, phonetics, linguistics, physiology, digital processing, information engineering, computer science, pattern recognition and artificial intelligence. How to make the speech recognition putting to practical uses is the hotspots of current research of speech recognition technology.This paper studies the speaker recognition, which is a branch of speech recognition and can be divided into speaker verification and speaker identification, it also studies transmission of the voice characteristics packet on the Internet. The first, this paper expounds the basic principle of speaker recognition, and then makes a detailed analysis of the entire process extracting the voice signal feature, as well as the various training models applying speech recognition, including the vector quantization using in this paper, Hidden Markov Models , Gaussian mixture model, artificial neural networks, support vector machines etc;In the speaker recognition systems, the height or low of identification rate directly determine whether it can makes for practical application. In order to improve the recognition rate, the paper lastly specialized in the study of the system in the distance measure, and improved on the distance measure and put forward the ways to weight the dimensions of the characteristics packet and optimize the weight value considering the aggregation extent in class and the distance extent between classes. In addition, the characteristics of packet transmission loss against Speaker Recognition System will have a greater impact on identification rate. This paper put forward the ways to cross the packet in the front packaging and substitute the lost packet in the service. Meanwhile, this paper achieved a distributed speaker recognition system, and practically proved that the proposed method in this paper is effective. Experiments show that it, by improving the method of measurement and handling characteristics of packet when the network environment is not good, increases the recognition rate of the system and makes the system with good recognition results.