| Summary: | Visual speech information plays an important role inlipreading under noisy conditions or for listeners with a hearing impairment. This research has developed several programming algorithm related to software development of lips pattern recognition in video imaging of 25 people for 17 Indonesian syllable pronunciation utilizing Hidden Markov Model (HMM) method.
Based on the results, lips motion pattern recognition for the same pattern of lips condition in database produces R = 0.77 of correlation coefficient, and for the different pattern of lips condition yields R = 0.38 of correlation coefficient. The best lips condition is shown for red lips which have R = 0.74 of correlation coefficient, whereas correlation coefficients for lips with mustache and pale are R = 0.68 and R = 0.38.
This research also analyzes the influence of syllable to lips pattern recognition. Bilabial syllable gives 77% of performance, 77% of dental, and 63% of palatal, while between /a/, /i/, /e/, /o/ phonemes, the best performance is obtained for /a/ phoneme which has no similarity with other phonemes. This research is very beneficial for other pattern processing researches in lips pattern recognition and health aids field.
|
|---|