一种基于噪声动态检测的语音端点检测算法

HMM技术在语音识别中得到较为成功的应用．本文基于HMM的孤立词语音识别系统的识别策略，提出了两个改进的算法．第一种算法是对背景噪声进行动态估计的汉语语音端点检测算法，较一些常规的端点检测方法如基于能量的端点检测方法，具有更高精度和鲁棒性，第二种算法针对分裂法进行码本设计时，可能出现除数为零的计算错误，提出了避免出现计算错误的改进算法．通过进行汉语孤立词语音识别实验，表明这两种算法是有效的，得到较高的识别率．

Endpoint detection algorithm of a speech based ondynamic noise estimating

Hidden markov model (HMM) is successfully used in speech-recognition.Based on the speech-recogniton of isolated-word with HMM,two algorithms are proposed in this paper.The first algorithm detects the beginning and ending boundaries of a speech by dynamic estimating environmental noises.This algorithm can work more accurately and robust than time-domain-energy-based endpoint detection.The second proposes a improvod algorithm,which can avaid errs in calculation when designing code book with splitting algorithm.The experiment results in isolated-word speech-recognition show thattwo algorithms contribute to the increase of recognition rate.