A study of data normalization measured by an electro-magnetic articulograph

Abstract

We investigated the normalization of the data measured using an electro-magnetic articulograph (EMA). The data normalization was needed because the size of the articulator of each subject is different from the sex and/or body size. Moreover, tongue movement range depends upon the speaking styles, clear or unclear. Our experiments were as follows: tongue tip movements during articulation of non-native consonant clusters were measured. Speakers were 2 Japanese and 2 Germans. Speech samples were four nonsense words, bnaht, pnaht, gnaht, knaht. Tongue tip displacement D (mm) and moving time T (ms) between first and second consonant in consonant clusters were measured. Dx (X component of D) was normalized by the difference between maximum value of X (Xmax) and minimum value of X (Xmin) in each utterance, i.e., Dnx = Dx/(Xmax−Xmin). Also, Dy (Y component of D) was normalized, i.e., Dny = Dy/(Ymax−Ymin). Hence, Dn (normalized D) was Dn = (Dnx2+Dny2)1/2. T was normalized by word length L (ms), i.e., Tn = T/L. Before the normalization, the measured data was localized by each speaker on the T-D plane, while the normalized data were not localized on the Tn−Dn plane. Accordingly, it was suggested that this simple normalization method would be effective in this experiment.