摘要： 针对多人说话改变点检测问题，提出一种改进的BIC说话人分割算法。采用固定窗BIC算法对音频流进行分割，利用基于递归的分割算法和变长窗口的BIC算法确认潜在的分割点。实验结果表明，与其他BIC算法相比，该算法的准确率、召回率和综合性能较高。

关键词: BIC准测, 广播音频分割, 准确率, 召回率

Abstract: To detect the voice change of speakers, an improved Bayesian Information Criterion(BIC) algorithm for speaker segmentation is proposed in this paper. The algorithm detects potential acoustic changes using fix-windows BIC algorithm, and validates the potential acoustic changes by two methods. One is the variable-size analysis window BIC algorithm, the other is based on the recursion of segmentation algorithm. Experimental result shows that the algorithm achieves better results, and compared with the BIC algorithm, the precision ratio, recall ratio and F-measure are improved.

Key words: Bayesian Information Criterion(BIC), broadcasting audio segmentation, precision ratio, recall ratio

中图分类号: 

• TP391