Speaker selecting arrangement and speaker adaptation model compilation device, speaker selective method, program and speaker adaptation model compilation program for speaker selection

摘要

The acoustic feature quantity of the speaker hourly changing, while corresponding to change, being accurate, try to be able to select the speaker where stably the acoustic feature quantity is close to the utterance speaker. Speaker score calculation expedient 22 calculates, the long haul speaker score which for example is based on optional several speeches (the respective degree of logarithm yu u of the plural speaker models which are remembered in the speaker model memory section 31 for the acoustic feature quantity) for example calculates brief it is based on the speech of brief speaker score. Long haul speaker selective expedient 23 selects the speaker who corresponds to the speaker model of the predetermined number whose long haul speaker score is high. Brief speaker selective expedient 24 selects the speaker who corresponds to the speaker model whose a quantity which is fewer than as much as predetermined numbers the speaker whom long haul speaker selective expedient 23 selects as an object, brief speaker score is high.