Nauman Dawalatabad
Research Scientist at Zoom Communications, San Jose, CA, USA
I specialize in developing Audio AI systems at scale.
Experienced in different areas like data processing, multilingual code-switched ASR, speaker diarization/verification, multimodal model, speech translation, source separation, TTS, and on-device models. Proven track record of translating research into production and open-source frameworks used across industry and academia. large language models, speaker diarization, and healthcare AI systems.
About
I am a Research Scientist at Zoom Communications working on production-scale Automatic Speech Recognition (ASR) for Voice AI Agents, healthcare speech systems, and word timing estimation.
Previously, I was a Postdoctoral Associate at the Massachusetts Institute of Technology (MIT CSAIL), where I worked with Dr. James Glass on robust speech recognition, trustworthy healthcare AI, uncertainty-driven pseudo-label filtering, and privacy-preserving voice deepfakes.
I completed my Ph.D. and M.S. in Computer Science at IIT Madras, working with Prof. C. Chandra Sekhar and Prof. Hema Murthy. My research focused on speaker diarization, continual transfer learning, and robust speech systems.
I have also contributed extensively to the open-source SpeechBrain toolkit as a core contributor during my internship at MILA, Montreal, Canada with Prof. Mirco Ravanelli and Prof. Yoshua Bengio.
Experience
- Production-grade ASR systems for Voice AI Agents in French and Hindi-English code-mixed settings.
- Zoom's first Healthcare-focused ASR systems for doctor-patient conversations using LLM persona based synthetic TTS data and added contextual biasing. Used for generating clinical notes.
- Lexicon-free multilingual word boundary estimation model used as core module for diarization, ASR, and EOU detection.
- Worked with Dr. James Glass on robust ASR and healthcare AI systems.
- Uncertainty-driven pseudo-label data filtering and calibration for speech recognition.
- Multimodal dementia detection from long-form interviews.
- Privacy-preserving voice deepfake generation for healthcare data augmentation.
- Selected as IEEE ICASSP Rising Stars in Signal Processing (2023).
- Worked on on-device streaming ASR model compression for Samsung Galaxy mobile phone devices.
- Led Conformer-T model compression ideas resulting in more than 60% model size reduction.
- Received Samsung E-Spot Award for impactful work.
- Core contributor to the SpeechBrain open-source toolkit.
- Developed ECAPA-TDNN based speaker diarization systems achieving state-of-the-art performance.
- Contributed models, clustering methods, and infrastructure for speaker recognition research.
Selected Publications
Skills
Some Invited Talks & Recognitions
- Judge: IEEE Signal Processing Cup Competition, IEEE ICASSP 2025.
- Invited Research Talk: Speech Research in Healthcare at Zoom, AI Developers, Zoom Headquarters, San Jose, 2024.
- Alumni Guest Invited Talk: Robust Speech Recognition, IIT Madras, 2023.
- Guest Lecturer: Speaker Recognition and Diarization, MIT 6.345, MIT CSAIL, 2022.
- IEEE ICASSP Rising Star in Signal Processing, 2023.
Featured in News
- CNBC News: Voice AI Cloning.
- Boston Globe News: Speech voice conversion.
- MIT News: Speech Anonymization.
- IEEE Spectrum AI News: Detect Deep Audio fakes.