Comparison of Three Method for Entering Tone Recognition
July 13, 2012
Posted by on
The entering tones are recognized using rule-based method, maximum entropy model and support vector machine. For those syllables which are affirmatively entering tone syllables, or undoubtedly non-entering tone syllables, rules are made use of to recognize. But for the ambiguity syllables, statistics-based method is utilized to disambiguate. Rules can reduce the recognition problem’s scale, and the statistical method can improve the accuracy of entering tone recognition. We compare the results of rule-based method, maximum entropy model and support vector machine. the precision using rules is the lowest and the precision using maximum entropy model and support vector machine are much higher than rule-based method. From the precision, the accuracy, F-measure and efficiency, maximum entropy model is better than support vector machine.
Full Paper in PDF Document