Phoneme-less Hierarchical Accent Classification

Lin, Xiaofan ; Simske, Steven

HP Development Company

关键词: speech recognition;    accent classification;    interactive voice response system;   

RP-ID  :  HPL-2004-166

学科分类：计算机科学（综合）

美国|英语

来源: HP Labs

【 摘 要 】

This paper introduces a novel accent classification method. Compared with existing methods, it has two unique features. First, it does not explicitly utilize phoneme information. Second, it is built on top of the gender classification. We have tested the proposed algorithm on datasets that are completely independent of training data. The accuracy of distinguishing American accent and British accent is 83%. We have also compared the accent classification with the gender classification in terms of accuracy and the saturation behavior with respect to length of utterance. Notes: Copyright IEEE. To be presented at and published in the 38th Annual Asilomar Conference on Signals, Systems, and Computers, 7-10 November 2004, Pacific Grove, CA 4 Pages