基于小波分解和信号相关函数的语音端点检测

SPEECH ENDPOINT DETECTION ALGORITHM USING WAVELET DECOMPOSITION AND SIGNAL-CORRELATION

为检测和分离含噪语音信号中的信号段和噪声段,提出一种基于小波分解和信号相关函数的检测方法。该方法对含噪信号进行多层小波分解,利用相邻层重构信号间的相似性,通过信号相关计算来检测语音端点。实验表明:该方法能够较准确地在噪声污染的音频中检测出语音端点,其抗噪声干扰能力强于美尔倒谱检测法。

In order to detect and separate speech from noisy audio signals,a new algorithm based on wavelet decomposition and signal-correlation is proposed.The algorithm makes multi-layer wavelet decomposition on noisy signal and takes advantage of the similarity of reconstructed signals in adjacent layers to detect end point of speech through signal-correlation computation.Experiments indicate that the algorithm can detect endpoint of speech from noise-contaminated audio signals precisely and present more robustness than the Mel Cepstrum detection algorithm in ability of anti-noisy interference.