噪声环境中基于HMM模型的语音信号端点检测方法

Speech Signal Endpoint Detection Method Based on HMM in Noise

在噪声环境下如何提高语音信号端点检测的准确性是自动语音识别（ＡＳＲ）研究中的一个重要课题．常用的基于短时能量的端点检测方法对于能量较低的音节或在信噪比较低的环境下，检测性能不够理想．讨论了一种基于ＨＭＭ模型的语音信号端点检测方法．先用训练的方法生成背景噪声和废料的模型，再用Ｖｉｔｅｒｂｉ解码算法对待测信号进行处理，并给出了具体的实现方法．实验测试结果表明，基于ＨＭＭ的端点检测方法的检测性能接近于人工检测，方法是有效的．

To improve performance of endpoints detection in noisy environments is a significant subject in automatic speech recognition (ASR). The performance of general endpoints detection methods based on short time energy is unsatisfied for lower energy syllable or in the environments with lower signal to noise ratio. The endpoints detection method based on HMM is discussed. The background model and garbage model are built with training, then the Viterbi decoding algorithm is used to process the speech signal. the steps of realization are presented in this paper. The results from experiments show that this method is very effective and the performance is approaching to that of manual detection.