子带特征在英语词重音检测中的应用(英文)  

Subband Features for English Lexical Stress Detection

作　　者：陈楠[1] 贺前华[1] 

机构地区：[1]华南理工大学电信学院,广州510641

出　　处：《科学技术与工程》2008年第2期494-498,共5页Science Technology and Engineering

基　　金：国家自然科学基金(60572141)资助

摘　　要：对于英语等"重音节拍语言",词重音是一个非常重要的韵律学特征、提出巴克谱子带能量(BSSE),巴克谱子带倒谱(BSSC)和巴克倒谱(BSCC)等三种使用子带方法,模拟人类听觉系统进行英语词重音检测的新特征。首先研究了在英语词重音检测中不同频带的贡献度,然后测试了在连续语音条件下使用这些特征的系统性能。试验结果表明,高频带对重音识别的贡献度比其它频带大,新特征的识别效果也比传统特征好。Lexical stress is an important prosodic feature, especially for stress-timed language such as English. Three novel features are proposes, which used subband method to simulate discrimination property of human auditory system, for English lexical stress detection. The proposed features are Bark Scale Subband Energy （BSSE）, Bark Scale Subband Cepstrum （BSSC） and Bark Scale Cepstral Coefficients （BSCC）. Firstly, the contributions of different frequency regions for stress detection are investigated, then the performances of these features are evaluated in continue sentences. Experiment results showed that the high frequency region is more valuable than the others, while the new features give significant improvement over traditional ones.

关 键 词：重音检测 子带分析 巴克谱 

分 类 号：TN912.3[电子电信—通信与信息系统]