Robust feature extraction for automatic speech recognition

语言可能是人类最自然、最有效的交流方式。因此，通过语音与机器交流一直是许多人的梦想。在过去的五十年里，自动语音识别领域取得了重大进展。虽然目前可用的语音识别器在无噪声的办公环境中工作得相当好，但由于背景噪声和其他失真的存在，当它们部署在现实环境中时，它们的性能会急剧下降。本课题将研究鲁棒语音识别问题。

Speech is perhaps the most natural and efficient mode of communication for humans. Therefore, it has always been a dream for many people to communicate with machines via speech. Significant advances have been made in the last five decades in the area of automatic speech recognition. Though the currently available speech recognisers work reasonably well in noise-free office environments, their performance deteriorates drastically when they are deployed in real-life situations due to the presence of background noise and other distortions. The problem of robust speech recognition will be researched in this project.