Robust Voice Activity Detector by combining sequentially trained Deep Neural Networks
S.M. Raufun Nahar, Atsuhiko Kai · 2016
Deep Neural Network (DNN) has been hot topic in the field of information processing recently. As a vast field of information processing, speech processing has also been aided by Deep Neural Network. After Deep Neural Network is introduced to speech processing, its performance has outperformed its older methods. But still there is room for improvement and researchers are working on to find new ways to make the fullest use of DNN. In the same direction, we have been carrying out our research to find improved results. Our work is about combining two DNNs to perform voice activity detection where one DNN has the role of Denoising Autoencoder (DAE) and the second one acts as Voice Activity Detector (VAD) and to train them sequentially to develop a robust voice activity detection system for noisy and reverberant speech through distant microphone in the real environment scenario like meeting room or so.