Exploring nonlinear relationships between speech face motion and tongue movements using Mutual Information

![CDATA[In the current study, we used Mutual Information (MI) to determine the amount of shared information between face motion and tongue movements during speech. We tracked face motion using a passive marker-based motion capture system and measured tongue motion employing Electromagnetic Articulography. The results show widespread associations between the two types of motion, albeit with predominantly relatively low MI values. More importantly with regard to practical applications, a pronounced speaker variability was observed. We further investigated the temporal frequency distribution of the shared information using a wavelet decomposition yielding one-octave band-limited subbands. It was found that the lowest frequency subband contained substantially higher amounts of shared information.]]