Souvik Ghosh

• Courses taken: Statistical Methods in AI, Digital Image Processing, Advanced NLP(LLMs), Computer Vision, Technology, Product and Entrepreneurship
• Research focus on multimodal AI, Speech Technologies, LLMs and Representation Learning.
• Working with CVIT Lab in the Audio Visual Team guided by Professor CV Jawahar and Professor Vinay Namboodiri

• Explored the beauty of interdisciplinary education.
• Final Year Thesis on IOT and Edge ML for patients with Epilepsy
• Major Projects in Harassment Detection and Women Safety. Runners up at Nasscom Lab 2 Market

A generic and efficient modular framework that adapts an existing frozen pre-trained Text-to-Speech model into a lip-synchronized speech generator.

A mobile app for silent video to speech communication leveraging a adapted VSR using a novel constrained beam search strategy

The first Single Stage Multi Domain OCR for Indian languages, showing better results than all Industry grade OCRs like that of Google, AWS, Azure and open souce SOTA models. Our commitment to open source remains and the model and codes have been made open source.

Explores Multimodal sentiment analysis using a novel ensemble early fusion deep learning architecture. Opens the door for targetted sentiment analysis in constrained environments.

With the advent of AI based content creation, clean and annotated datasets on Indian languages are necessary. This work proposes a dataset of balanced make and female speakers for Indian languages.