摘要 |
PROBLEM TO BE SOLVED: To provide a characteristic word learning device for business section extraction for learning a characteristic word enabling business section extraction of good precision with a small calculation quantity, and a business section extraction device which extracts a business section using a learning result thereof.SOLUTION: A recognition result word unit conversion unit of a characteristic word learning device for business section extraction receives a learning text and a speech recognition result, generates a speech recognition dictionary and a language model from long-unit-mode words obtained by subjecting the learning text to long-unit morpheme analysis, generates a word unit conversion table between short-word-mode words obtained by subjecting the learning text to short-unit morpheme analysis, and the long-unit-mode words, converts the short-word-mode words into long-unit-mode words using the word unit conversion table, and outputs the long-unit-mode words. A characteristic word extraction unit extracts characteristic words (long-unit-mode word) appearing one-sidedly only in a typical utterance. A business section extraction device extracts a business section using the characteristic words and a typical utterance average appearance position. |