Automatic Classification of Palatal and Pharyngeal Wall Shape Categories from Speech Acoustics and Inverted Articulatory Signals

Inter-speaker variability is pervasive in speech, and the abil-ity to predict sources of inter-speaker variability from acoustics can afford scientific and technological advantages. An impor-tant source of this variability is vocal tract morphology. This work proposes a statistical model-based approach to classify-ing the shape of the hard palate and the pharyngeal wall from speech audio. We used principal component analysis for the parameterization of the morphological shape. Analysis using K-means clustering showed that both the palate and the pharyn-geal wall shape data group into two major categories. These in turn are used as targets for automatic classification using acous-tic features derived at the utterance level with OpenSmile and at the model level using GMM based posterior probability super-vectors. Since articulatory motions are dependent on morpho-logical shape, the model uses estimated articulatory features on top of speech acoustics for improving the classification perfor-mance. Experimental results showed 70 % and 63 % unweighted accuracy for binary classifications of palate and pharyngeal wall shapes in the rtMRI database, respectively, and 63 % for the palate shape on the X-Ray Microbeam database. Index Terms: speech production, vocal tract morphology, acoustic-to-articulatory inversion, speaker recognitio