Speaker selecting device, speaker adaptive model creating device, speaker selecting method, speaker selecting program, and speaker adaptive model making program

摘要: To enable selection of a speaker, the acoustic feature value which is similar to that an utterance with accuracy and stability, while adapting changes even when speaker every moment. A score calculating means (22) calculates long-time (log likelihood each plurality models stored in model storage section (31) respect value) based on arbitrary number utterances, for example, short-time utterance, example. selecting 23 selects speakers corresponding predetermined having high score. 24 models, smaller than sore high, from among selected by 23.