Sound source enhancement device, sound source emphasis learning device, sound source enhancement method and program

摘要

PROBLEM TO BE SOLVED: To provide a sound source enhancement technique using real number frequency conversion with different analysis lengths based on deep learning.;SOLUTION: An sound source enhancement device in which x_t is an observation signal of the t-th block and φ_t is an acoustic feature of the t-th block obtained by extracting from the observation signal x_t, comprises: an attack determination vector generation unit which generates from the acoustic feature φ_t an attack determination vector a_t which is a vector indicating a determination result of whether or not the t-th block is an attack; a window function vector generation unit which generates from the attack determination vector a_t a window function vector z_t; a j-th output sound generation unit which generates a j-th output sound s^_j,t^C from the observation signal x_t and the acoustic feature φ_t by using a calculation unit corresponding to the j-th window function; and an output sound generation unit which generates from the j-th output sound s^_j,t^C and the window function vector z_t an output sound s^_t emphasizing a target sound included in the observation signal x_t.;SELECTED DRAWING: Figure 5;COPYRIGHT: (C)2020,JPO&INPIT