I am a Senior Research Scientist at Netflix Research & Globalization DSE, working on multimodal and multilingual LLMs for speech understanding, reasoning, and generation, with the goal of removing language barriers from the Netflix experience.
Previously, I was a Senior Research Scientist at Hume AI, building the expressive speech generation engine behind its first empathic voice interface, and an Applied Scientist on the AWS AI Speech team. I received my Ph.D. in Electrical Engineering from the Signal Analysis and Interpretation Lab (SAIL) at USC, advised by Panayiotis Georgiou and Shrikanth Narayanan.
馃寪 Website 路 馃帗 Google Scholar 路 馃捈 LinkedIn
- Speech generation: post-training for expressive and controllable TTS and cross-lingual voice
- Speech recognition and understanding: multilingual long-form ASR and LLM-based translation
- Evaluation: diagnostic audio and multimodal LLM judges informed by human preferences
- Earlier: speaker verification, affective computing, and emotion recognition
Recent (2026)
- SpeechCritic: Learning a Diagnostic Speech Judge from Limited Human Preferences 路 code
- Region-Aware Masking for Accent-Robust Cross-Lingual Text-to-Speech
- Louder, Longer, Livelier: Acoustic Shortcuts and Underspecified Rationales in Speech LLM Judges
- Align Then Reason: A Multimodal Lip-Sync Judge for Dubbing
Earlier
- Self-Supervised Speaker Verification with Simple Siamese Network and Self-Supervised Regularization 路 ICASSP 2022
- Speaker-Invariant Affective Representation Learning via Adversarial Training 路 ICASSP 2020
- Linking Emotions to Behaviors through Deep Transfer Learning 路 PeerJ CS 2020 路 code
More on my website and Google Scholar.
Speech and audio ML, evaluating speech models with LLM judges, and the path from a Ph.D. in affective computing to production speech systems.