Transcribing and Translating the Tamil Nadu Government Videos using Whisper AI

The growing demand for multilingual translation systems in diverse social and governmental contexts has driven research into efficient and accurate speech-to-text pipelines. This paper introduces a robust framework for Tamil-to-English translation using a fine-tuned Whisper AI model, complemented by linguistic structure analysis through N-gram frequency modeling. The system begins with speech input in Tamil, which is transcribed and translated using the Whisper architecture optimized over nine training epochs. The translated outputs are then analyzed using N-gram techniques to quantify linguistic complexity and semantic richness across multiple departments. The model demonstrates high translation accuracy, low validation loss, and the ability to adapt across varying domains such as Finance, Art & Culture, Tamil Nadu Assembly, Education, and Social Welfare. Through the N-gram analysis, department-specific language usage patterns were identified, providing actionable insights into domain relevance and textual density. This dual approach of translation and structural analysis ensures contextual precision and highlights the model's capacity to support large-scale multilingual communication systems in public service and accessibility applications