Literary and Colloquial Dialect Identification for Tamil using Acoustic Features

TL;DR

The paper tackles Tamil dialect identification by distinguishing Literary Tamil (LT) and Colloquial Tamil (CT) using acoustic features, aiming to provide a front-end ADI for robust Tamil ASR without relying on transcription or language models. It adopts a data-efficient approach based on Gaussian Mixture Models (GMM) with MFCC features, avoiding language-specific annotations, and introduces a dedicated LT-CT speech corpus along with text data. A key empirical finding is that vowel nasalization differences between LT and CT support classification, yielding up to approximately 88% accuracy with 256 mixture components, while balancing computational cost. The work offers a practical, adaptable framework for dialect identification that can extend to other languages and dialects, reducing dependence on annotated data and language-dependent resources, and it suggests future work in deeper nasalization analyses and potential transcription-based comparisons.

Abstract

The evolution and diversity of a language is evident from it's various dialects. If the various dialects are not addressed in technological advancements like automatic speech recognition and speech synthesis, there is a chance that these dialects may disappear. Speech technology plays a role in preserving various dialects of a language from going extinct. In order to build a full fledged automatic speech recognition system that addresses various dialects, an Automatic Dialect Identification (ADI) system acting as the front end is required. This is similar to how language identification systems act as front ends to automatic speech recognition systems that handle multiple languages. The current work proposes a way to identify two popular and broadly classified Tamil dialects, namely literary and colloquial Tamil. Acoustical characteristics rather than phonetics and phonotactics are used, alleviating the requirement of language-dependant linguistic tools. Hence one major advantage of the proposed method is that it does not require an annotated corpus, hence it can be easily adapted to other languages. Gaussian Mixture Models (GMM) using Mel Frequency Cepstral Coefficient (MFCC) features are used to perform the classification task. The experiments yielded an error rate of 12%. Vowel nasalization, as being the reason for this good performance, is discussed. The number of mixture models for the GMM is varied and the performance is analysed.

Figures (2)

• Figure 1: Examples showing differences between literary and colloquial Tamil
• Figure 2: Comparison of the degree of nasalization of vowel 'aa' in literary and colloquial Tamil