关键词: 小波子带能量, BIC准测, 广播音频分割, 准确率, 召回率

Abstract: Based on wavelet sub-band average-energy variance and Bayesian information criterion，audio segmentation algorithm is proposed，for the sliding variable-size analysis window BIC algorithm suffers from a large amount of redundancy change points.The approaches detect acoustic changes by partitioning a continuous audio stream into sub-segment using wavelet sub-band average-energy variance，and then detect acoustic changes by improved sliding variable-size analysis window BIC algorithm in sub-segment.The experiment shows that this approaches have achieved a better results，and compared with the sliding variable-size analysis window BIC algorithm，this algorithms have improved the precision，recall and F-measure.

Key words: wavelet sub-band energy, Bayesian Information Criterion（BIC）, broadcasting segmentation, recall, precision