Korean Grapheme-to-phoneme Conversion Algorithm Based on Decision Tree

In the resource construction of Korean speech information processing, grapheme-to-phoneme conversion technology plays a vital role. According to the Korean pronunciation rules and the law of word boundaries phonetic variation, this paper defines the attributes used to construct the decision tree, and realizes the Korean phonetic conversion algorithm based on the CART decision tree, so that the phonetic representation can reflect the phonetic changes such as syllable weakening, falling off, increasing and alienation in the continuous speech stream, and accurately obtain the phonemes corresponding to the graphemes. By the cross-validation, the average conversion accuracy can reach to 94.63%.