| 摘要 |
PROBLEM TO BE SOLVED: To provide a feature amount compensation apparatus etc. capable of calculating a feature amount of voice with high accuracy by selecting smooth noise environment which is close to change of actual noise environment. SOLUTION: The feature amount compensation apparatus comprises: a feature extraction section 102 for extracting a feature vector of voice in each frame of a plurality of frames in which input voice is divided for each predetermined period; a noise environment sequence estimation section 103 for estimating a noise environment sequence which is a sequence of a plurality of noise environment sequences for generating each feature amount vector, based on reversion for each noise environment for indicating probability that a feature vector sequence which is a plurality of feature vectors for each extracted frame, and the feature vector of a current frame and at least a preceding frame or a following frame of the current frame, is generated in noise environment; a compensation vector calculation section 104 for calculating a compensation vector corresponding to each noise environment of the estimated noise environment sequence; and a feature vector compensation section 105 for compensating the feature vector based on the compensation vector. COPYRIGHT: (C)2008,JPO&INPIT
|
