Jinyu Li

Also known as Jin-Yu Li, Li Jinyu

Jinyu Li is a Partner Applied Science Manager at Microsoft in Redmond, where he leads a science team working on speech, translation and language technologies. Trained in electrical engineering at the University of Science and Technology of China, he completed a PhD at the Georgia Institute of Technology in 2008 and joined Microsoft in 2008. His work runs from compact deep neural networks for speech recognition on small devices in 2014 and the 2015 book Robust Automatic Speech Recognition to the WavLM speech model, the VALL-E speech synthesizer and Phi-4-multimodal. The IEEE named him a Fellow in 2025 for deep-learning-based speech technology innovation and commercialization, and the International Speech Communication Association made him a Fellow in 2026.

Now at: Microsoft
Works: 315
Citations: 13,334
h-index: 53
Badges: Rising
Bridge
Loyal

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

1997–2008

Degrees at USTC and Georgia Tech

He earns B.E. and M.E. degrees in electrical engineering and information systems from the University of Science and Technology of China in 1997 and 2000. In 2004, he wins Georgia Tech's Colonel Oscar P. Cleaver Award for the highest score on the ECE Ph.D. preliminary exam, and he receives an Interspeech 2006 Best Student Paper award. He completes his PhD in electrical and computer engineering at the Georgia Institute of Technology in 2008.
2008–2014

Compact speech models at Microsoft

He joins Microsoft in Redmond, Washington, in 2008. In 2014, he and Rui Zhao, Jui-Ting Huang and Yifan Gong present an Interspeech paper that trains a small deep neural network to match the output distribution of a large one, for devices with limited computing and storage. He also receives a Microsoft research technical transfer award in 2014.
2015

A book on robust speech recognition

In October 2015, he publishes Robust Automatic Speech Recognition: A Bridge to Practical Applications with Li Deng, Reinhold Haeb-Umbach and Yifan Gong. The book surveys noise- and reverberation-robust techniques for speech recognizers built on Gaussian mixture models and deep neural networks.
2021

WavLM and an end-to-end survey

In October 2021, he coauthors WavLM, a self-supervised model that learns masked speech prediction and denoising for full-stack speech tasks, with its training data scaled to 94,000 hours. In November 2021, he posts his sole-authored survey Recent Advances in End-to-End Automatic Speech Recognition, which APSIPA Transactions on Signal and Information Processing accepts. APSIPA names him an Industrial Distinguished Leader in 2021.
2023

VALL-E and the Furui Prize

In January 2023, he is one of 13 authors of VALL-E, a neural codec language model that synthesizes personalized speech from a 3-second recording of an unseen speaker. His end-to-end survey wins the APSIPA Sadaoki Furui Prize Paper Award in November 2023.
2025

IEEE Fellow and Phi-4-multimodal

The IEEE elevates him to Fellow in January 2025 for contributions to deep-learning-based speech technology innovation and commercialization. In March 2025, Microsoft releases Phi-4-multimodal, which handles speech, vision and text in a single model, and he is a coauthor of its technical report. The WavLM paper receives an IEEE Signal Processing Society best paper award in December 2025.
2026

ISCA Fellow and committee vice chair

He becomes vice chair of the IEEE Speech and Language Processing Technical Committee in January 2026. In 2026, the International Speech Communication Association names him a Fellow for deep-learning-based speech technology innovation, adoption and commercialization.
As of October 2026

Leading speech science at Microsoft

He is a Partner Applied Science Manager at Microsoft in Redmond, leading a science team on speech modeling, translation and language technologies, with research on speech-LLMs and end-to-end speech recognition and translation. In June 2026, his team's Azure-Speech-05-2026 model takes first place on both the main and multilingual OpenASR leaderboards.