Sawit Kasuriya, Takatoshi Jitsuhiro, Genichiro Kikui
Thai Speech Recognition by Acoustic Models
Mapped from Japanese
Abstract:This report describes Thai speech recognition by using Japanese acoustic models.
There was no Thai speech database for use in this study, and it takes a lot of time to
correct a large number of speech database entries and label them. We also have only a
small amount of Thai speech data, which was not alignment data. Therefore, we made
initial models by mapping another language's acoustic models, and trained the initial
models by using a small amount of data. We used phoneme mapping by the IPA table
and made the initial Thai models from Japanese models. In our experiments, the initial
models were trained by using isolated-word utterances by eleven Thai speakers. We
evaluated the performance of isolated-word recognition with context-independent and
context-dependent models.