Acoustic feature quantity conversion apparatus, acoustic model adaptation apparatus, acoustic feature quantity conversion method, and program

摘要

PROBLEM TO BE SOLVED: To improve a recognition rate even when learning data is not sufficiently available to obtain.SOLUTION: An acoustic feature value extraction unit 12 extracts an acoustic feature value series from a sound signal including an object sound phenomenon. A phoneme label addition unit 13 adds a phoneme label to the acoustic feature value series. An object label addition unit 14 adds an object label to the acoustic feature value series. A conversion model generation unit 15 learns a conversion model that converts an acoustic feature value of other than the object sound phenomenon into an acoustic feature value of the object sound phenomenon. A pseudo-acoustic feature value generation unit 16 generates a pseudo-acoustic feature value series in which the acoustic feature value of other than the object sound phenomenon out of the acoustic feature value series is converted into the acoustic feature value of the object sound phenomenon using the conversion model. An acoustic model learning unit 17 learns an acoustic model using the acoustic feature value series and the pseudo-acoustic feature value series.SELECTED DRAWING: Figure 1