Research Of Speech Endpoint Detection Based On Neural Network

Speech endpoint detection is an important process of speech analysis, speech recognition and speech monitor. The purpose of the system is to determine the input speech signal, and find out the starting point or end point. So we can storaging and processing effectivt voice signals, reduce data processing time and computation.Even though the technology of speech endpoint detection has obtained good precision in the condition of undisturbed, but in the practical application, the system capability would decline remarkably because of the influence of noise, some even can not work. Therefore, how to improve endpoint detection system's robustness and increasing the stability of the system is ours direction.First this paper introduces several typical speech endpoint detection method, and introduces a speech endpoint detection method which based on RBF neuarl networks, uses wavelet analysis to obtain features as the input of neural network, made some improvements to RBF neuarl networks, and uses colligate discriminant criteria to improve the result. The experiment result shows the flexibility and the real-time performance of the algorithm,and this method obtains satisfying effect with common noise environments.