Electro-Magnetic Articulography data stabilization for speech synchronized articulatory animation

Abstract

To produce speech synchronized articulatory animation, Electro-Magnetic Articulography (EMA) data is one type of important training data for establishing the relationship between speech and articulatory movements. Because the EMA data is easily contaminated by the head motion during the capturing process, this paper proposes a real-time robust stabilization system for EMA noisy data. Firstly, global motion parameters are obtained by fitting the EMA noisy data between the reference frame and current frame with random sample consensus algorithm. Secondly, multiple evaluation criteria, i.e., global motion parameters and location errors of corresponding EMA noisy data matches, are fused by an adaptive low-pass filter to smooth global motion for obtaining correction vector. Finally, motion compensation is applied to the current frame by using correction vector, and stabilized EMA data is obtained. By comparing between the EMA noisy data and stabilized EMA data, the experimental results demonstrate the system can increase the average peak signal-to-noise ratio around 5.92 dB, the perceptive comfort on the speech synchronized articulatory animation driven by the stabilized EMA data.