A Thai Speech Translation System for Medical Dialogs

2. Speech Recognition The language adaptation techniques developed in our lab [5] enables us to rapidly bootstrap a speech recognition system in a new target language given very limited amount of training data. The Thailand’s National Electronics and Technology Center gave us the permission to use their Thai speech data collected in the hotel reservation domain. They provided us with a 6 hours text and speech database recorded from native Thai speakers. We divided the data into three speaker disjoint sets, 34 speakers were used for training, 4 speakers for development, and another 4 speakers for evaluation. The provided transcriptions were manually pre-segmented and given in Thai script. We transformed the Thai script into a Roman script representation by concatenating the phoneme representation of the Thai word given in the pronunciation dictionary. The motivation for this romanization step was threefold: (1) it makes it easier for non-Thai researchers to work with the Roman representation like in the grammar development, (2) the romanized output basically provides the pronunciation which makes things easier for the speech synthesis component, and (3) our speech engine currently does not handle Thai characters.