UMD CS Ph.D. Student Receives NVIDIA Graduate Fellowship 

Imagine an AI system that doesn’t just transcribe speech but can truly understand, reason and respond like an expert. This vision is at the heart of University of Maryland Ph.D. student Sreyan Ghosh’s research, which seeks to revolutionize audio processing in artificial intelligence. His work, which addresses limitations in data efficiency and representation learning, has earned him the highly competitive NVIDIA 2025-2026 Graduate Fellowship, positioning him to advance what AI systems can achieve with sound. Ghosh is among ten recipients selected from nearly 600 applications worldwide. 

The fellowship, which provides up to $60,000 in funding and a high-end NVIDIA GPU, also includes an in-person summer internship at NVIDIA. This opportunity will enable Ghosh to collaborate with leading experts and access cutting-edge technology.

“This award inspires me to push the boundaries of innovation and address the most critical challenges in the field,” Ghosh said. “Winning the NVIDIA Ph.D. Fellowship is an incredible honor and a recognition of the potential impact of my research in advancing audio processing.”

Ghosh’s research, conducted under Distinguished University Professor Dinesh Manocha and Professor Ramani Duraiswami in the Gamma Lab, focuses on addressing foundational limitations in audio processing and developing AI systems capable of expert-level audio understanding and reasoning. His work leverages data-efficient deep learning, representation learning and innovative synthetic data pipelines to enhance audio perception and reasoning capabilities in AI systems.

The scarcity of high-quality labeled audio data poses a significant challenge in audio processing. To overcome this, Ghosh is exploring advanced neural architectures and synthetic data pipelines to develop general-purpose audio understanding models. His work aims to enable AI systems to comprehend complex audio contexts, generate interleaved audio-text responses and perform multi-audio reasoning.

“Audio processing is integral to creating accessible and intelligent AI systems that can better understand and interact with the world,” Ghosh explained. He highlighted potential applications for his research, including improving voice assistants, enhancing educational tools and advancing audio-based diagnostics in healthcare. By addressing data limitations, he aims to make audio-based AI technologies more widely accessible across industries and communities.

Looking ahead, he plans to design scalable audio encoding architectures capable of handling diverse audio inputs, from short sounds to extended conversations. His long-term vision includes developing full-duplex conversational AI agents with expert-level reasoning capabilities. He believes these innovations could transform accessibility, education and interactive technologies.

The NVIDIA Graduate Fellowship Program, now in its 24th cycle, is a cornerstone of the company’s efforts to support academic research. It seeks to identify and invest in promising Ph.D. students working on topics that contribute to advances in accelerated computing, artificial intelligence, robotics and autonomous systems. Recipients are selected based on academic achievements, faculty nominations and the relevance of their research to NVIDIA’s mission of driving innovation in these fields.

—Story by Samuel Malede Zewdu, CS Communications