基于可信度变化趋势的音频分割算法

提出基于可信度变化趋势的音频分割算法。采用定长滑动窗检测结构减少累积错误，在窗内计算各音频帧的可信度，根据可信度的变化趋势检测跳变点，以避免阈值选择和硬门限判决造成的误检。实验结果表明，该算法的分割性能优于基于KL2距离、基于隐马尔可夫模型、基于贝叶斯信息准则和基于熵变化趋势的音频分割算法。

Audio Segmentation Algorithm Based on Trend of Believable Degree Change

This paper proposes an audio segmentation algorithm based on trend of Believable Degree(BD) detection. It uses the detection structure of fixed-size sliding window to avoid accumulative errors. BD of every audio frame is computed in sliding window, and jump points are detected according to the trend of BD changes, so that detection errors due to threshold setting and hard threshold judging can be avoided. Experimental results demonstrate that the algorithm is superior to KL2-based algorithm, HMM-based algorithm, BIC(Bayesian Information Criterion)-based algorithm and Entropy-based algorithm.