Investigating self-supervised architectures for learning speech representations

Under guidance of Prof. Ganesh Ramakrishnan

In this project I along with my research group have explored various audio encoders.

The findings of the project has helped improve state-of-the-art results on multi-modal video captioning.

Responsibilities :

• Explored various pre-training techniques for learning audio embeddings.
• Used PASE architecture to better handle low resource setting with rich audio features
• Work got accepted in Interspeech 2020
• Technologies: Python and PyTorch