中文摘要：

为解决由乐器弹奏的音频信号只能依赖人工分割的问题,针对音频信号的短时平稳性,提出一种以短时方差稳定性度量为评判标准的音频分割算法。采用分层结构遍历音频流,通过判断定长窗口内每个音频点左右两端信号的方差稳定性,检测音频跳变点。该算法避免了因积累误差和硬门限判决造成的误检,能解决传统算法中不同乐器发声需要设置不同参数的问题,能准确查找到分割点。实验结果表明,该算法计算速度快,对由乐器产生的音频有较好的分割效果。

英文摘要：

To deal with the dependence of musical instrumental audio segmentation on manpower,an audio segmentation method based on the short-time stationarity of musical instrument signals was proposed.By using the hierarchical structure with fixed length windows,the audio stream was travelled and the stationarity at both sides of the signal was analyzed to detect the segment points.This method can avoid the mistake detections caused by accumulated errors and hard thresholds,which can solve the difficulties of various parameters with different musical instrument and improve the accurate rate.The experimental simulation on musical instrument verifies that the method is feasible,fast and reliable.