鲁棒语音识别技术综述  

Review of robust speech recognition

作　　者：吕钊[1] 吴小培[1] 张超[1] 

机构地区：[1]安徽大学计算智能与信号处理教育部重点实验室,安徽合肥230039

出　　处：《安徽大学学报（自然科学版）》2013年第5期17-24,共8页Journal of Anhui University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61271352);安徽大学校学术与技术带头人引进工程基金资助项目(02303203)

摘　　要：鲁棒语音识别是为了解决噪声环境所引起的语音识别系统识别和训练不匹配的情况.依据噪声对语音识别系统的影响,从信号空间、特征空间及模型空间3个层面上分别对语音增强技术、特征增强技术及语音模型补偿、增强技术进行了总结,并分析了不同方法的特点、实现及应用.To solve the mismatch between the training and recognition environment, some robust speech recognition methods were proposed. Based on the influence of noise on Automatic Speech Recognition （ASR） system, some classified and summarized robust speech recognition technologies in the aspects of speech enhancement, feature enhancement and model compensation ／ enhancement aiming at the signals space, feature space and model space of ASR system were presented in this paper. Furthermore, some main ideas of these approaches were analyzed.

关 键 词：鲁棒 语音识别 语音增强 特征增强 语音模型补偿 增强 

分 类 号：TN912.34[电子电信—通信与信息系统]