基于语音拖音段的端点检测算法研究

为了进行有效的语音信号处理,并降低语音信号的冗余度,通常采用端点检测技术来提取语音信号中的有效部分。本文在传统语音端点检测方法的基础上,提出了一种基于基音周期对语音段末尾进行判别的方法,针对汉语发音都是以浊音结尾的特点,同时利用基音周期对浊音段信号比较敏感这一特性,能够有效地避开汉语语音信号尾部拖音段中所包含的无效信息,既提高了端点检测的准确性,又减少了后续语音识别系统样本训练时间。实验结果证明,该方法对于汉语中孤立词末尾的拖音段,可以得到较好的端点检测效果。

Research on Speech Endpoint Detection Based on Slow Tone

The valid information of the voice signal is often extracted by endpoint technology for the sake of increasing the efficiency and decreasing the redundancy of the voice signal processing.Based on the traditional endpoint detection method,the paper proposes the pitch detection method in order to improve the accuracy of the endpoint detection.This new method of the voice endpoint detection can remove the invalid information by using the pitch's sensitivity to the sonant signal in view of considering all the Chinese pronunciation is end by vowel.Not only the accuracy of the follow-up voice recognition can be improved but also the sample training time can be reduced.Experimental results demonstrate that this new method has the better performance especially for the slow tone of the isolate word.