Effects of Corpus Choice on Statistical Articulatory Modeling

ABSTRACT: Statistical articulatory modeling is based on the analysis of a collected speech corpus to extract parameters that should be able to resynthesize the articulatory movements in the original corpus. As the parameters are based directly on the corpus, the size and content of the corpus have important effects on the model. In this paper we compare statistical articulatory models based on three corpora of different sorts: one rather exhaustive consisting of sentences, one limitied, but articulatory balanced and one limited and more unbalanced. The articulatory data was collected in a simultaneous recording with the electromagnetic articulograph Movetrack and the infrared stereo capture system Qualisys. The subject was a female native speaker of Swedish. The comparison of the different corpora focused on the articulatory space covered, the outcome of the statistical component analysis, articulatory speed and hyperarticulation.