GENERATING LARGE UNITS OF GRAPHONEMES WITH MUTUAL INFORMATION CRITERION FOR LETTER TO SOUND CONVERSION

摘要

Words are provided a method and apparatus for segmentation into constituents. According to the invention, mutual information scores are determined for pairs of graphoneme units found in a set of words. Gras each graphoneme unit to include at least one character. Graphoneme units of one pair of graphoneme units are combined based on the mutual information score. This results in a new graphoneme unit is formed. According to an aspect of the invention, the n- gram syllable model is trained based on words in the segment into syllables using mutual information. N- gram syllable model is used to segment a phonetic representation of a new word into syllables. Similarly, a list of morphemes is formed using mutual information, new words stemming n- grams that can be used to segment a sequence of morphemes is training.