复杂噪声环境中的语音端点检测

Speech endpoint detection in real noise environments

提出了一种适用于复杂加性噪声环境的语音端点检测方法。通过对噪声类型的归纳,建立自适应的平稳噪声模型,并根据模型搜索信号能量非平稳的区域。然后基于浊音在频域上的谐波结构,在此区域内检测浊音,从而排除非平稳噪声的干扰。最后根据信号能量精确搜索语音起止点。与目前典型端点检测算法的对比实验表明,在大多数复杂噪声环境下,该算法具有较好的准确率。

A method of speech endpoint detection in environments of complicated additive noise is presented. Based on the analysis of noise, an adaptive model of stable noise is proposed to detect the section where the signal is unstable. Then the voice is detected in this section by its harmonic structure, and the accurate endpoint is searched using energy. Compared with the typical algorithms, this algorithm operates reliably in most real noise environments.