Ultrasound-Based Silent Speech Interface Using Convolutional and Recurrent Neural Networks

Silent Speech Interface (SSI) is a technology with the goal of synthesizing speech from articulatory motion. A Deep Neural Network based SSI using ultrasound images of the tongue as input signals and spectral coefficients of a vocoder as target parameters are proposed. Several deep learning models, such as a baseline Feed-forward, and a combination of Convolutional and Recurrent Neural Networks are presented and discussed. A pre-processing step using a Deep Convolutional AutoEncoder was also studied. According to the experimental results, an architecture based on a CNN and bidirectional LSTM layers has shown the best objective and subjective results.