MASK MAKING DEVICE, NOISE SPECTRUM ESTIMATING DEVICE, AND SPEECH RECOGNITION DEVICE

摘要

PROBLEM TO BE SOLVED: To provide a mask making device capable of correctly estimating band speech likeliness by referencing time spectrum patterns for the time over a future frame from a past frame.;SOLUTION: A speech input section 1 is input with a speech signal and generates an analysis frame. A band analysis section 2 determines the band power from the speech signal of the analysis frame generated by the speech input section 1. A delay memory 3 memorizes the time series of the band power determined from the band analysis section 2. A time length search section 61 determines the time length of the pattern based on the time series of the band power. A time length memory 62 memorizes the time length. A speech likeness estimation section 4 estimates the speech likeness of the band in the present frame, and produces a mask output 5 based on the speech likeness.;COPYRIGHT: (C)2007,JPO&INPIT