Improved tone model for low complexity tone recognition

This paper proposes an improved tone model for low complexity tone recognition. This model recognizes tone by using only estimated vowel signals, called vowel magnitude difference function, vowel-MDF (V_MDF) so that we can reduce the number of input frames and the fundamental frequency (F₀) negative influence from neighboring syllables in continuous speech. Vowel segmentation is therefore an important part of the tone recognition. The tone model improves vowel segmentation by estimating vowel parts in the voiced/unvoiced detection process more precisely. This results in more accurate and less complex tone recognition. The proposed method is tested using 44 Thai words, selected from voice activation for GPS systems and phone dialing options. The experimental results show a 3.03% improvement in recognition accuracy, compared with the conventional vowel segmentation.