THE CU-MFEC CORPUS FOR THAI AND ENGLISH SPELLING SPEECH RECOGNITION

Much of the efficiency of any Automatic Speech Recognition (ASR) system depends on its speech corpus. This is even more so for recognizers designed for specific tasks. Naturally, an ASR for spelling recognition performs better if it is trained with a spelling speech corpus rather than a generic one. Although several speech corpora are available in Thai, we are still lack of Thai spelling speech corpora. This paper reports collection of experiences gained from constructing CU-MFEC, a Thai spelling speech corpus designed for form filling or other applications of similar nature. CU-MFEC corpus employed 100 speakers and encompassed 58 hours and 10 minutes of speech. There are four sets of the corpus; Alphabets with short pauses, Continuous free spelling, Sentences, and Numbers and commands. We evaluated its efficiency by utilizing CU-MFEC with speech recognition tasks and found the accuracy rate of 79.37% for spelling task and 54.92% for connected spelling task.