摘要
说话人分割聚类是近几年新兴起的语音信号处理研究方向,它主要研究如何确定连续语流中多说话人起止时间的位置,并标出每个语音段对应的说话人。这项研究对自动语音识别、多说话人识别和基于内容的音频分析等都具有重要的意义。根据说话人分割和聚类实现过程不同,本文从异步策略和同步策略的角度回顾了十年来国内外研究的主流算法、技术和代表系统,对比了不同代表系统在近几年NIST富信息转写评测的结果,最后讨论了目前还存在的问题,并对未来的发展进行了展望。
Speaker segmentation and clustering, which are focused on the determination of the starting and ending time points in multi-speaker audio flows and labeling the speech signal segments with labels corresponding to the identity speaker, have gradually become a hotspot in the field of speech signal processing in the recent years. It plays an important role in auto- matic speech recognition (ASR), multi-speaker recognition and content-based audio signals analysis. Based on the different implementation processes used in the speaker segmentation and clustering, this paper gives a detailed review of the state-of-the- art algorithms, techniques and typical systems proposed in the past decade from the aspects of asynchronous and synchronous strategies. And the performances of the typical systems are compared through the NIST Rich Transcription (RT) evaluations in recent years. The existing problems are discussed and the future prospects of this research area are also described at the end.
出处
《信号处理》
CSCD
北大核心
2013年第9期1190-1199,共10页
Journal of Signal Processing
基金
北京市教育委员会科技发展计划重点项目(KZ201110005005)
国家自然科学基金资助项目(61072089)
关键词
说话人分割聚类
异步策略
同步策略
贝叶斯信息判决
Speaker segmentation and clustering
Asynchronous strategy
Synchronous strategy
Bayesian information criterion
作者简介
马勇男,1977年生,江苏新沂,北京工业大学博士研究生,主要研究方向为语音信号处理和模式识别。E—mail:may773@emals.bjut.edu.cn
鲍长春男,1965年生,内蒙古赤峰,博士,北京工业大学教授、博士生导师,IEEE信号处理学会会员,国际语音通信学会(ISCA)会员,中国电子学会理事,中国声学学会理事,信号处理学会委员,《通信学报》编委会副主任委员、《信号处理》和《数据采集与处理》编委。主要研究方向为语音与音频信号处理。E—mail:baochch@bjut.edu.cn