Speech Tracking Algorithm for an Automated Reading Tutor in Filipino
Nicole Anne A. Palafox, Rhandley D. Cajote · 2019
Speech tracking stimulates continuity of reading, improves miscue detection and provides immediate feedback. Automated reading tutors (ART) need this feature to emulate coaching of human teachers. A previously developed ART for Filipino (Tanglaw) uses an automatic speech recognizer (ASR) to implement a miscue detection feature, but it does not track the current reading position. In this research, experiments on implementing speech tracking for Filipino ART to indicate the word to be read, in order to mimic guided reading. The next word will be highlighted when the previous word has been uttered correctly. Results show that using a phoneme-level ASR with online decoding, dynamic programming with revised cost function, and speaker's reading rate using words-correct-per-minute (WCPM) as parameters yield the highest timing accuracy of tracking (TAT) of 41.18% compared to 29.03% and 37.11 % with Levenshtein distance at most one and half word recognized thresholds, respectively.