APPARATUS AND METHOD FOR DETECTING VOICE BASED ON MULTILAYER STRUCTURE OF DEEP NEURAL NETWORK AND RECURRENT NEURAL NETWORK

摘要

The voice recognition performance may be improved by exactly detecting a section in which a voice signal exists in an audio signal including a noise signal using a model combining a deep neural network and a recurrent neural network. An apparatus for detecting a voice may receive an audio signal including a noise signal and voice signal, extract a feature vector for each frame from the audio signal, and perform an operation for determining a VAD label corresponding to the voice signal or the noise signal as a result of processing the feature vector extracted from the frame using a model combining a deep neural network and a recurrent neural network.;COPYRIGHT KIPO 2020