Signal processing device, signal processing method, and non-transitory computer readable recording medium storing signal processing program

Abstract

A signal processing device includes: a mixed feature amount conversion part that converts a mixed acoustic signal including a plurality of acoustic signals into a mixed feature amount; a mask estimation part that estimates a plurality of masks based on the mixed feature amount; an acoustic signal conversion part that converts a plurality of separated feature amounts calculated using the plurality of masks into a plurality of separated acoustic signals; an environment acoustic signal extraction part that extracts, as an environment acoustic signal, the mixed acoustic signal in the environment sound section estimated based on the plurality of separated acoustic signals from the mixed acoustic signal; and an environment sound feature amount conversion part that converts the environment acoustic signal into an environment sound feature amount. The mask estimation part estimates the plurality of masks based on the mixed feature amount weighted using the environment sound feature amount.