Noise-Robust Speech Recognition Based on Multiple kernel Learning

摘要：

To improve noise-robust speech recognition system. In this paper a multi-class classification was presented which based on multi-kernel learning and multi-resolution wavelet coefficients of extracting of features parameters. According to multi-kernel learning theory, multi-class classifier was constructed by the binary tree combined strategy. Using Mel frequency cepstral coefficient (MFCC) based on wavelet transform, an anti-noise speech recognition system was implement of Chinese isolated words, speaker- unspecified and medium vocabulary. Experimental results show that in speech recognition the proposed method has good recognition rate, promising anti-noise capability and robustness.

关键词： Support vector machine multi-kernel learning Gaussian kernel multi—class classification speech recognition.wavelet transform feature extraction

作者: Qiu Shuxiong Li Zhishu Zhang Lei Sun Yafei Wang Di

作者单位: The College of Computer Science, Sichuan University, Chengdu, China The Archives of Sichuan University, Chengdu, China

会议类型: 国际会议

会议名称: 2010 International Conference on Information Security and Artificial Intelligence(2010年信息安全与人工智能国际会议 ISAI 2010)

会议地点: 成都

会议语种:英文

页码: 1402-1405

在线出版日期: 2010-12-17（万方平台首次上网日期，不代表论文的发表时间）