people

I am a PhD student at the Department of Electrical Engineering, Indian Institute of Science (IISc), Bangalore, working with Dr. Sriram Ganapathy as part of the LEAP Lab. I also serve as an Assistant Professor in the Department of Electronics and Communication Engineering, College of Engineering Trivandrum.

Research interests

My work centers on representation learning for audio and speech, and on building small, efficient audio language models — including their end-to-end pre-training. This is complemented by a broader interest in information theory and communication, and grounded in a strong mathematical foundation in probability theory, linear algebra, and calculus for deep learning.

I have carried out complete self-supervised pre-training of audio foundation models such as HuBERT and SSAST. Two projects that reflect this work:

• SmallALM — a lightweight, 135M-parameter Audio Language Model that pairs a frozen self-supervised audio encoder with a compact language model (SmolLM2-135M) through a lightweight projector, aimed at showing that competitive audio understanding doesn’t require billion-parameter models.
• SSAST-MLM — a re-formulation of SSAST pre-training that replaces its two original self-supervised losses (masked patch classification and generation) with a single, unified masked-prediction objective in the style of HuBERT/BERT, turning spectrogram pre-training into a masked “language” modeling task over quantized audio tokens.

See the CV, publications, and projects pages for more details.