| 摘要 |
PROBLEM TO BE SOLVED: To individually learn the feature spectral envelope structure and a feature microstructure of sound, when a three component feature which is constituted of a spectral envelope feature, a spectral microstructure feature and a power time envelope feature are extracted from a multiple sound signal. SOLUTION: A time sequence of a power spectrum of an input sound signal is output (step S1), and the initial value of the three component feature which is constituted of the spectral envelope feature, the spectral microstructure feature and the power time envelope feature is generated. Updated value of the three component feature is calculated so that a power spectrum time sequence model, obtained by an updated three-component feature becomes similar to the time sequence of the power spectrum of the sound signal, under prescribed restricting conditions regarding three components (step S2); and it is determined as to whether repetition of three component feature updating process satisfies a prescribed standard. If the standard is not satisfied, the three-component feature updating process is performed (step S4); and if the standard is determined as being satisfied, the updates value of the three-component feature is output (step S5). COPYRIGHT: (C)2009,JPO&INPIT
|
