Speech reconstruction using a deep partially supervised neural network

Abstract

Statistical speech reconstruction for larynx-related dysphonia has achieved good performance using Gaussian mixture models and, more recently, restricted Boltzmann machine arrays; however, deep neural network (DNN)-based systems have been hampered by the limited amount of training data available from individual voice-loss patients. The authors propose a novel DNN structure that allows a partially supervised training approach on spectral features from smaller data sets, yielding very good results compared with the current state-of-the-art.

Keywords

• speech processing
• medical signal processing
• medical disorders
• Boltzmann machines
• statistical speech reconstruction
• deep partially supervised neural network
• larynx related dysphonia
• Gaussian mixture models
• restricted Boltzmann machine arrays
• voice-loss patients
• DNN structure
• partially supervised training approach