摘要 |
PROBLEM TO BE SOLVED: To provide a pronunciation evaluating method of evaluating properties of a speech such as accuracy of a pronunciation of a known sentence by a speaker, a phoneme series model learning method, a device using their methods, a program and a recording medium. SOLUTION: The speaker is made to read the predetermined sentence and then acoustic scores are obtained by using a speech recognition technique. The obtained acoustic scores are put together by collections of one or more phonemes, and weighted addition of the acoustic scores is performed to find a score of a phoneme series. An array of scores of a plurality of phoneme series constituting the predetermined sentence is compared with a phoneme series model (an array of scores of previously learnt phoneme series) to select the closest learning data. Evaluation data corresponding to the selected learning data are outputted by using a database (DB) in which previously learnt data and evaluation data (marks, advices, comments, etc.) are related to each other. COPYRIGHT: (C)2007,JPO&INPIT
|