METHOD AND APPARATUS FOR COMBINED LEARNING USING FEATURE ENHANCEMENT BASED ON DEEP NEURAL NETWORK AND MODIFIED LOSS FUNCTION FOR SPEAKER RECOGNITION ROBUST TO NOISY ENVIRONMENTS

摘要

A method and apparatus for joint learning using deep neural network-based feature enhancement and a modified loss function for robust speaker recognition in a noisy environment are presented. A deep neural network-based feature reinforcement and combined learning method using a modified loss function for speaker recognition robust to a noisy environment according to an embodiment receives a voice signal and uses at least a beamforming algorithm and a reverberation removal algorithm using a deep neural network. a pre-processing step of learning to remove noise or reverberation components using any one or more; a speaker embedding step of learning to classify a speaker from the speech signal from which noise or reverberation components are removed using a speaker embedding model based on a deep neural network; and connecting the deep neural network model included in at least one of the beamforming algorithm and the reverberation removal algorithm with the deep neural network-based speaker embedding model for speaker embedding, and then performing joint learning using a loss function. can be done