Comparison of DCT and autoencoder-based features for DNN-HMM multimodal silent speech recognition | IEEE Conference Publication | IEEE Xplore