Speech Feature-based Machine Learning and Deep Learning Approach for Assamese Dialect Identification

Dialect identification helps identify various dialects within a certain language. The Assamese language is spoken widely across Assam, India, and its neighboring states. Identifying dialects within a language is important in gaining insights into its social, cultural, and historical context. The broadly recognized dialects in Assamese include Central, Lower, and Upper Assamese dialects. Recent developments in Machine Learning (ML) and Deep Learning (DL) have shown promising results in dialect identification in various languages, rather than relying on traditional methods that are labor intensive and require manual intervention. Features of speech signals play a vital role in their respective task. This paper explores the use of Random Forest and Long Short-Term Memory (LSTM) models for the automatic identification of Assamese dialects. By leveraging our proposed dataset of three dialects, extracting relevant features such as Mel-Frequency Cepstral Coefficients (MFCCs), Energy, Formants (F1, F2, F3), and Spectral Flux, and ML/DL architectures, this study contributes to the development of automatic dialect identification of Assamese dialects. The results showed that both the Random Forest and LSTM models achieved an accuracy of 97%, indicating good performance with limited availability of data.