Multifractal Detrended Cross-Correlation Analysis for Characterization of Spoken Language – A New Method to explore the Genesis of Languages

• DOI: 10.52783/jns.v14.2100
• Publication Date: 2025-03-17T08:32:07Z
• Authors: Suparna Panchanan; Nilav Darsan Mukh...; Ranjan Banerjee; Most Mahabuba Islam; Shankha Sanyal; Dipak Ghosh; Debmalya Mukherjee

This work presents a novel use of chaos-based non-linear techniques coupled with statistical methods for spoken language characterization in the speech signal domain. Our goal is to create a framework that highlights linguistic commonalities. While Multifractal Detrended Cross-Correlation Analysis (MFDXA) assesses long-term cross-correlations between languages using the cross-correlation coefficient as a measure of similarity, Multifractal Detrended Fluctuation Analysis (MF-DFA) is used to examine linguistic correlations among the languages. Bengali, Assamese, Maithili, Odia, Nepali, Manipuri, Hindi, Urdu, Marathi, Gujarati, Punjabi, Konkani, Tamil, Telugu, Malayalam, Kannada, and Sanskrit are among the seventeen Indian languages that are the emphasis of the present work. The speech corpus, which includes speakers of both sexes, is mostly composed of unplanned conversational material on a variety of subjects, including social welfare, agriculture, and in-person interviews. This model is unique in that it avoids the conventional use of linguistic information. Our findings reveal notable deviations from established linguistic theories in cases such as Bengali-Gujarati, Hindi-Tamil, and Bengali-Kannada, indicating that current language classifications may benefit from re-evaluation. Statistical tools like ANOVA and Mahalanobis Distance have been used to validate the current study. The proposed method offers a valuable approach to investigating the origins of languages within a global framework