VAKTA-SETU: A Speech-to-Speech Machine Translation Service in Select Indic Languages
We present a deployment-ready Speech-to-Speech Machine Translation (SSMT) system for English-Hindi, English-Marathi, and Hindi-Marathi language pairs. Our SSMT pipeline integrates ASR, Disfluency Correction (DC), Machine Translation (MT), and TTS to enable seamless spoken language translation. Additionally, we develop a Text-to-Text Machine Translation (TTMT) service for six translation directions, leveraging a LaBSE-based corpus filtering tool to improve training data quality. Our system serves government initiatives, tourists, the Indian judiciary, and farmers, addressing real-world multilingual challenges. We publicly release our datasets, models, and insights gathered from large-scale stakeholder demonstrations.