Embedded Files
I am a Senior Machine Learning Engineer at Meta, specializing in Audio, Speech, and On-Device AI. My expertise lies in bridging the gap between Generative AI and production-scale efficiency—designing models that are both generatively powerful and computationally grounded.
My work consistently leverages Self-Supervised Learning (SSL), Flow Matching, and GANs to build robust representations. While my primary focus is Generative Speech Restoration, I have extensive experience applying these techniques to Computer Vision (insect-scale robotics) and NLP (malware detection/action recognition), allowing me to solve complex, multimodal problems.
Current & Past Impact
Current & Past Impact
At Meta: Leading the implementation of Generative Audio Restoration models and real-time quality assessment for the Meta Family of Apps, impacting over 3 Billion Daily Active Users.
At Samsung Research (Bixby): Developed production-grade Wake Word detection pipelines using Knowledge Distillation and Few-Shot Learning for the Galaxy ecosystem.
Academic Roots: Research at the University of Washington’s AIR Lab, where I custom-built hardware and low-computation Computer Vision models for insect-scale robotics.
Recent Publications
Recent Publications
Balasubramanian, S., Jimenez Amador, J.A., Kalgaonkar, K., Hor, K.W. and Srinivasan, S., 2025. "SMARTMOS: Modeling Subjective Audio Quality Evaluation for Real-Time Applications." In Proc. Interspeech 2025 (pp. 3165-3169).
Focus: Bridging the gap between subjective human perception and automated metrics for real-time audio.
Balasubramanian, S., Jajodia, A., Srinivasan, G., 2021. "Towards noise robust trigger-word detection with contrastive learning pre-task for fast on-boarding of new trigger-words." arXiv:2111.03971 [cs.SD]. [Link]
Focus: Leveraging contrastive learning (SSL) to accelerate the deployment of new wake words.
Balasubramanian, S., Chukewad, Y. M., James, J. M., Barrows, G. L., Fuller, S. B., 2018. "An Insect-Sized Robot that uses a Custom-Built Onboard Camera and a Neural Network to Classify and Respond to Visual Input." International Conference on Biomedical Robotics and Biomechatronics (Biorob). [Link]
Focus: Hardware-software co-design for resource-constrained vision in robotics.
Patents
Patents
Sivakumar Balasubramanian, et al. "System and method for accent-agnostic frame-level wake word detection." U.S. Patent No. 12,272,357, issued April 2025.
Sivakumar Balasubramanian, et al. "Method of generating a trigger word detection model, and an apparatus for the same." U.S. Patent No. 12,236,939, issued February 2025.
Sivakumar Balasubramanian, et al. "System and method for command fulfillment without wake word." U.S. Patent App. 17/937,198, published January 2024.
Google Sites
Report abuse