Natarajan Balaji Shankar

Hi! I’m a Ph.D. candidate in the Electrical and Computer Engineering department at UCLA, where I conduct my research at the Speech Processing and Auditory Perception Laboratory under the guidance of Prof. Abeer Alwan.

My research lies at the intersection of machine learning and speech processing, with a focus on building robust automatic speech recognition (ASR) systems for low-resource domains such as children’s speech and African American English. Recent work introduces Selective Attention Merging, a model-merging method that sets state-of-the-art Word Error Rate on the MyST child speech corpus, and CHSER, a dataset and case study on generative speech error correction tailored to child ASR. Earlier work spans speech-only domain adaptation, non-autoregressive ASR architectures, automated assessment of children’s oral language abilities, and dialect density estimation for African American English. I’m broadly interested in adaptation strategies for untranscribed speech, adapting speech foundation models to new domains, identifying and addressing biases in language models used in educational applications, and building equitable spoken language assessment systems for children.

In summer 2025, I interned at Qualcomm’s Audio & ML Research group in San Diego, where I worked on neural speech coding methods that balance speech quality and bitrate efficiency for deployment in resource-constrained and noisy real-world conditions.

Prior to UCLA, I earned my undergraduate degree in Electronics and Communication Engineering from the National Institute of Technology, Tiruchirappalli. There, I worked in the Signal and Image Processing Laboratory under Prof. Varun Gopi on preprocessing techniques for color fundus images, supporting improved aberration detection and disease classification.