用于语音端点检测的鲁棒性特征提取新方法

针对实际噪声环境中的语音端点检测问题,提出了一种适用于不同噪声类型的鲁棒性特征提取方法。该方法把基音检测中的循环平均幅度差函数应用到端点检测的特征提取中,并与基本的谱熵相结合,具有适用范围广和不需要噪声先验知识的优点。仿真实验验证结果表明:该特征对于多种类型的噪声有明显的抑制作用,并且在低信噪比时仍然有效。

New Robust Feature Extraction Method for Speech Endpoint Detection

In order to solve the problem of speech endpoint detection in real-world noisy environments, a new robust feature extraction technique for speech endpoint detection was proposed. In the technique, the cyclic average magnitude difference function in the fundemental tone detection in combination with the basic spectral entropy was applied to extract the features of speech endpoint. The proposed technique is characterized by adaptation to the various noisy environments without any need for a prior knowledge of the noise. Simulation results show that the feature extraction is effective to reduce the impact of various kinds of noise under low signal-to-noise ratio.