阈值自适应语音自动分割系统模型  

Speech automatic segmentation system modeling with adaptive threshold adjustment

作　　者：张俊星[1] 石立新[1] 王都生[1] 

机构地区：[1]大连民族学院机电信息工程分院,辽宁大连116600

出　　处：《计算机工程与设计》2010年第8期1886-1888,F0003,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(60573124)

摘　　要：针对有声出版物语音分割系统,提出了一种阈值自适应加相似度判决的系统分割模型,基于脚本中的先验知识提出了能量阈值自适应分割算法。对于传统的端点检测算法无法排除的干扰,为提高系统的抗干扰能力以增强其适用性,提出了基于语音单元相似性进行分析判决的新方法。测试结果表明,无干扰时,系统分割的正确率100%,每个语音文件包含两个人为干扰信号时,系统分割正确率98.8%,能够满足有声出版物语音自动分割的需要。To achieve the design of speech automatic segmentation of talking publication, the system model with adaptive threshold and similarity detection capabilities is established. The adaptive energy threshold adjustment algorithm is put forward based on the prior knowledge of script. Aimed at the problem of interference which can’t be eliminated by the traditional endpoint detection algorithm, in order to improve the anti-interference ability and enhance its applicability, a new algorithm based on the similarity detection of speech units is provided. The test results show that the accuracy of system segmentation is 100% when the system without interference, but the accuracy of system segmentation is 98.8% when added two interference signals to each speech files on average. Practical application shows that the system meet the automatic segmentation requirements.

关 键 词：有声出版物 语音分割 阈值调整 自适应算法 相似性分析 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]