Multilingual Speech and Text Translation for Indian Regional Languages

In a multilingual country like India Language barrier between communities present significant challenges for effective cross-lingual communication. The development of robust translation systems that handle both text and speech across these languages is essential for fostering inclusivity and accessibility. In order to address these challenges this study aims to the exploration of multilingual speech and text translation using the model, a state-of-the-art transformer-based model tailored for Indian languages. The proposed application leverages IndicTrans2’s capabilities to seamlessly translate text and spoken content between 22 Indian languages, enhancing accessibility and inclusivity. This solution is specifically designed to bridge communication barriers for speakers of different Indian languages, offering real-time translations that support both speech-to-text and text-to-speech functionalities and make our code available on GitHub. The system architecture, underlying model mechanics, training methodologies, and evaluation metrics are thoroughly discussed in paper. Future work aims to expand the model’s capabilities by incorporating additional dialects and optimizing real-time speech translation performance.