在线无监督说话人检索中稳健的模型自举算法-软件学报.PDF

ISSN 1000-9825, CODEN RUXUEW E-mail: jos@ Journal of Software, Vol.18, No.3, March 2007, pp .608−616 http://www.j DOI: 10. 1360/jos1 80608 Tel/Fax : +86-10 © 2007 by Journal of Software. All rights reserved. ∗ 在线无监督说话人检索中稳健的模型自举算法 + 付中华 , 张艳宁 (西北工业大学 计算机学院,陕西 西安 710072) A Robust Bootstrapping Algorithm of Speaker Models for On-Line Unsupervised Speaker Indexing + FU Zhong-Hua , ZHANG Yan-Ning (School of Computer Science, Northwestern Polytechnical University, Xi’an 710072, China) + Corresponding author: Phn: +86-29 E-mail: mailfzh@ Fu ZH, Zhang YN. A robust bootstrapping algorithm of speaker models for on-line unsupervised speaker indexing. Journal of Software, 2007,18(3):608−616. /1000-9825/18/608.htm Abstract: A robust bootstrapping framework, which employs Multi-EigenSpace modeling technique based on regression class (RC-MES) to build speaker models with sparse data, and a short-segments clustering to prevent the too short segments from influencing bootstrapping, are proposed in this paper. For a real discussion archived with a total duration of 8 hours, the significant robustness of the proposed method is demonstrated, which not only improves the speaker change detection performance but also outperforms the conventional bootstrapping methods, even if the average bootstrapping segment duration is less than 5 seconds. Key words: speaker indexing; speaker model; regression class; eigenvoice 摘 要: 基于回归树模型的多特征空间建模方法在回归类内部进行特征音分析,较好地解决了训练数据不足时说 话人模型的训练问题, 而短语音段聚类策略又进一步避免了过短的语音片断对自举训练的影响.验证实验采用了实 际录制的近8 小时的不同谈话数据.结果显示,即使平均自举片断长度小于5 秒,新方法依然非常稳健,不仅提高了说 话人改变检测的效果,而