Research of Pre-Processing about Uyghur in Southern Xinjiang for Speech Recognition

This paper analyzed the characteristics of Uyghur speech, and collected Xinjiang Uyghur Hetian dialect on the laboratory environment. At first, we made first order digital filtering pre-emphasis on samples, then applied double threshold endpoint detection by the combination of short-time average magnitude and short-time zero crossing rate as well as used Matlab simulation experiment to pick up 462 speech samples from 500 samples(Abstract precision rate was 92.4%). This paper will be the reference to the research on Uyghur speech recognition in Xinjiang.