带噪语流中浊音端点检测

浊音端点检测在语音处理中占有重要地位,在语音编解码、语音识别、语音增强处理中都需要用到端点检测技术。常规的以短时能量、过零率等作为判决特征参数的方法无法在低信噪比的系统中满足应用需求。本文以信号的共振峰和基音周期检测为基础检测浊音端点,算法首先提取语音信号的第一共振峰以及基音周期信息并以此为判决依据判断浊音的起点和终点。实验证明,这种方法在噪声环境中相对于传统的基于能量检测的或AMR_WB标准中的端点检测算法有更高的正确率。

Endpoint detection of noisy speech by the use of formant and pitch

Voice activity detection, which is needed in speech codec, speech recognition and speech enhancement, is very important in speech processing. Conventional voice activity detection methods based on some simple features such as short term energy cannot meet the demand of application in noisy environment. Speech formants and pitches are used as the features to detect the voice activity in this paper. The information of the first formant and the pitch of speech signal are used to detect the starting points and end points of active voice. Experimental results show that this method can obtain higher accuracy than ordinary detection methods based on energy and the method proposed in AMR_WB standard.