Fundamentals of speech recognition
Fundamentals of speech recognition
Look-Ahead Techniques for Fast Beam Search
ICASSP '97 Proceedings of the 1997 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP '97)-Volume 3 - Volume 3
Hi-index | 0.00 |
This paper presents a Thai syllable speech recognition system with the capability to achieve high accuracy of Thai syllable speech and Thai tone recognition. The recognition accuracy of 97.84% is achieved for Thai syllable speech recognition using the Continuous Density Hidden Markov Model (CDHMM). To provide a faster response, a beam pruning technique is applied, in which the result shows that by using this technique with an appropriate beam width, the recognition time can be reduced by more than 4 times. As Thai is tonal language, tone recognition is crucial for distinguishing meanings of Thai syllables. To obtain high rates of tone recognition in the Thai language, the CDHMM and a mixed acoustic feature method are employed. The tone recognition rates of 97.88%, 97.36%, 98.81%, 90.67% and 100.0% are achieved for mid, low, falling, high and rising tones, respectively.