一种基于能量聚类分析的句子语音端点检测法

针对语音复读系统等背景噪声相对较小且稳定的实际应用环境,提出一种改进的基于时域分析的句子语音端点检测算法。因为在此类应用环境中,对句子语音端点检测的干扰因素较少,且一般需要实现快速的句子语音端点检测。因此,简化了所要使用的语音特征参数,不再使用时域分析中常用的过零率,仅使用语音信号能量特征值辅以聚类分析完成语音端点检测。实验表明,本算法简化了端点检测的过程,可以便捷有效地检测出句子中的语音端点。

A Sentential Endpoint Detection Algorithm Based on Energy Eigenvalue and Clustering Analysis

To actual application system of low noise,such as voice replayer,an improved sentential endpoint detection algorithm based on time domain analysis was discussed in this paper.Because of less disturbed inspects in this application environment,and needing realize fast speed endpoint detection of sentences,simplify the parameters of voice character and do not use cross-zero rate.By using speech energy eigenvalue and clustering analysis,the endpoint of sentences can be detected very rapidly.Experiments show that the algorithm can simplify the process of sentential endpoint detection and have excellent performance.