The Acoustical and Language Modeling Issues on Uyghur Speech Recognition

Automatic speech recognition(ASR) is an important application that enables interaction of human beings with machines. Neural network based frameworks have been making some breakthroughs in low-source data processing so that the minority languages meet the great chance, in some cases. Thereupon, low-source languages, such as Uyghur, also get promising results in both speech recognition and natural language processing, as a result, a lot of papers have reported, one after another. In this paper, to begin with, several primary Uyghur speech recognition approaches are listed, then, their pros and cons are discussed respectively. we also attempt to discuss acoustic unit selecting for Uyghur speech recognition system, simultaneously n-gram and RNN language models are compared, in the end.