Effects of Rate of Articulation in Rhythm Formant Analysis-based Dialect Classification

This study explores the application of Rhythm Formant Analysis (RFA) in the categorization of dialects. To conduct this research, we focused on two distinct Assamese dialects, which are spoken by native speakers from the Jorhat and Nalbari districts of Assam. The study utilized speech data obtained from a total of 6 male and 4 female speakers each from Jorhat and Nalbari districts, respectively. These speech utterances served as the basis for extracting rhythm features, which were subsequently employed in generating dendrograms obtained. Dendrograms are a hierarchical clustering method used to group data based on their similarities. The evaluation of the clustering from the dendrograms has been done applying a majority vote rule. Notably, this study revealed a tendency for utterances of the same dialect to cluster together. Furthermore, this research also examined the impact of the speakers’ articulation rate during the second recording session. Two ablation studies were conducted to determine whether rhythm formants are sensitive to variations in the rate of articulation. The findings from these ablation studies yielded negative results, indicating that rhythm formants do not exhibit sensitivity to changes in the articulation rate. In linguistically diverse regions, the suggested method shows great potential for a range of applications in the fields of speech recognition and language identification besides dialect classification.