鲁棒性语音识别技术研究综述

Research review on robust speech recognition technology

鲁棒性语音识别是为了解决噪声环境和混响环境等外界因素所引起的语音识别系统训练和识别不匹配的情况,针对在噪声和混响条件下进行鲁棒性语音识别的问题,对现有的鲁棒性语音识别研究进行了总结,阐述了语音识别的主要流程和整体框架;从特征提取和声学建模两个方面对语音增强技术、语音分离技术进行介绍;分析鲁棒性语音识别技术主要难点和实现过程。最后对鲁棒性语音识别技术进行总结和展望。

Robustness of speech recognition is to solve noise and reverberation environment caused by the external factors such as speech recognition training and does not match the situation,under the condition of noise and reverberation for robustness of speech recognition problem,the robustness of speech recognition study of existing are summarized,this paper expounds the voice recognition of the main process and the overall framework,Speech enhancement technology and speech separation technology are introduced from two aspects:feature extraction and acoustic modeling.The main difficulties and implementation process of robust speech recognition technology are analyzed.Finally,the robust speech recognition technology is summarized and prospected.