A Study on the Appropriateness of Thai Words and Testing the Speech Recognition Technology Technique for Read-Aloud of Grade One Students in Thailand

Abstract

This research aimed 1) to collect and group Thai words and 2) to test the Speech Recognition Technology (SRT) technique for reading aloud. The research methodology was a mixed method which consisted of two stages belong to research aims: The first stage was a qualitative method to demonstrate the collection of Thai words for Grade One students by using a checklist table in 11 word categories to verify the details of Thai words and analyses by three Thai language experts, and the second stage was a quantitative method to test the SRT techniques in sound extraction by using the spectrogram sound feature extraction method and Mel Frequency Cepstral Coefficients (MFCC). The results showed that in stage 1, 2,423 Thai words were appropriate for Grade One students in11 categories; 1) single vowel-long sound, 2) single vowel-short sound, 3) spelled correctly according to the final consonant, 4) spelled incorrectly according to the final consonant, 5) tone marks, 6) no tone marks, 7) stable vowel, 8) transforming vowel, 9) reduced vowel, 10) words with diphthongs, and 11) words with leading consonants. stage 2, the ((MFCC) sound feature extraction method provided higher accuracy than the spectrogram. In future research, the researcher will apply the approach developed in this study to voice extraction, aiming to create an application that detects Thai reading aloud for Grade One students. This will enable testing of a wider variety of words and expand the target group in the future.