| 摘要 |
PROBLEM TO BE SOLVED: To provide an acoustic model generating device capable of efficiently carrying out acoustic model learning.SOLUTION: The acoustic model generating device of this invention comprises: a label generating voice recognition unit; a data selection unit; and an acoustic model learning unit. The label generating voice recognition unit performs voice recognition on input voice signals with reference to a language model and a base acoustic model to apply labels to voice recognition results of the voice signals and to output their reliability degrees and acoustic likelihood degrees. On the basis of input of the voice signals output by the label generating voice recognition unit, the labels, the reliability degrees, and the acoustic likelihood degrees, the data selection unit selects voice signals whose reliability degrees are greater than a reliability threshold and whose acoustic likelihood degrees are smaller than a likelihood threshold. Then, the acoustic model learning unit makes the voice signals selected by the data selection unit learn the base acoustic model to generate a learnt acoustic model. |
