Research On Mongolian Speech Recognition

This article analyses the problems related to the Mongolian speech recognition using the theory and methods of experimental phonetics and computational linguistics, such as design of building the corpus for Mongolian speech recognition, building the lexicon for Mongolian speech recognition, developing the automatic phonetic notation program, building the acoustic model and so on.The mainly achievement of this article are:1. Building the corpus for Mongolian speech recognition including3299sentence (approximate30hours) and the transcripts for it.2. Building the lexicon for Mongolian speech recognition including more than25k items, which is the essential part not only of automatic phonetic notation but also Mongolian speech recognition.3. Developing the automatic phonetic notation program, which can automatic transcript the written Mongolian to the phonetic symbol online. There is no one have done this before.4. Building the basic Mongolian speech recognition system.5. Analysing the factors that affect the recognition rate from the Mongolian phonetics features. Proposing the solution to improve the recognition rate.6. Prompting the difficulties of building Mongolian large vocabulary speech recognition system.