Abhinav Jain is an experienced Speech Recognition Researcher specializing in Deep Learning and ASR systems. He has conducted extensive R&D at Samsung R&D, focusing on Bixby Voice Intelligence and transitioning to Transformer-based E2E models. His expertise includes working with low-resource, accented, and multilingual speech recognition challenges.
Experience
Lead Engineer
Samsung R&D
Speech Recognition R&D for Bixby Voice Intelligence team. Successfully launching homebrewed Indian English Acoustic Model for all Bixby enabled devices in Asia. Involved in transitioning from hybrid ASR systems to Transformer based E2E Models using Tensor2Tensor. Application of SOTA techniques like Multi-condition Training, Simulated Reverberation, Speed perturbation, VTLP in Production Models. Honored with 3 Spot Awards and special MD Incentive Award for excellent performance.
Voice Intelligence R&D
Samsung R&D
Active research in Accented, Low-resource, Multilingual and Code-Switched Speech Recognition. Proposed a Mixture of Experts based Acoustic Model in hybrid ASR for Accented Speech Recognition. Mentoring multiple college student teams as part of Samsung PRISM program. 2nd rank in IITM Hindi ASR Challenge held in 2020.
Researcher
Microsoft Research
Applied Transfer Learning and Multitask Learning resulting in 25% relative improvement.
Education
IIT Bombay
M.Tech.
Computer Science
SKIT, Jaipur
B.Tech.
Information Technology