语音端点检测中判决机制的研究

Study on Decision Mechanisms for Speech Endpoint Detection

以短时能量和平均过零率为端点检测的参量,分别讨论了基于单门限判决、双门限-两级判决和双门限-三态转换判决机制进行端点检测的方法。实验结果表明,后两种判决机制都能比较准确地检测出语音端点。

This paper presents speech endpoint detection methods based on the single threshold decision, dual - threshold - and - dual - step and dual - threshold - and - three - state - converted decision mechanisms respectively, with the short - time energy and the average zero - crossing- rate being the endpoint detection parameters. Experiments show that the two latter mechanisms can more accurately detect the speech endpoint.