Sign In

Duc Le

Senior Staff Research Scientist, GenAI at Meta

Duc Le is a Senior Staff Research Scientist at Meta, specializing in large language models (LLMs) and audio processing. He has a strong background in automatic speech recognition (ASR) and acoustic modeling, having previously contributed significantly to Meta's AI Speech organization. Duc holds a Ph.D. from Purdue University, where he focused on advanced methodologies in speech processing.

Career Highlights::

  • Current Role: Senior Staff Research Scientist at Meta, focusing on LLMs and audio technologies.
  • Previous Experience: Duc has worked as a Staff Research Scientist at Facebook AI, where he played a key role in developing ASR technologies.
  • Education: He earned his Ph.D. from Purdue University in 2021.

Duc Le's expertise and contributions to the field of AI and speech technology have made him a notable figure in his area of research, and he continues to push the boundaries of what is possible with audio and language technologies at Meta.12

Highlights

Mar 26 · linkedin

I've had a blast at TikTok working on music ML and navigating the crazy world of generative AI over the past year. TikTok is a place where remarkable progress can be made in an incredibly short amount of time. I have learned and grown so much technically here, perhaps the most I ever did throughout my career.

For my next chapter, I will re-join Meta as a Senior Staff Research Scientist. More detail to come regarding what I'll be working on, likely somewhere at the intersection of audio understanding and generation. Thrilled to be back to help solve new and exciting challenges!

May 24 · linkedin

I will be attending ICASSP 2023 in Greece from June 4-10. Looking forward to meeting and catching up with friends and colleagues!

A few work to be presented at the conference that I was involved in (all done while at Meta):

  • Factorized Blank Thresholding for Improved Runtime Efficiency of Neural Transducers (https://lnkd.in/g2J8uw9c): A simple but effective method to drastically reduce the RTF and power consumption of RNN-T without any WER degradation - with Frank Seide, Yuhao Wang, Yang Li, et al.

  • Massively Multilingual ASR on 70 Languages: Tokenization, Architecture, and Generalization Capabilities (https://lnkd.in/gFwJYkgu): Methods to scale up multilingual ASR training to 70 languages, with excellent performance on out-of-domain datasets - with Andros Tjandra, Nayan Singhal, Abdel-rahman Mohamed, Ozlem Kalinli, et al.

  • Learning ASR Pathways: A Sparse Multilingual ASR Model (https://lnkd.in/gvVnXZmv): A method that takes advantage of multilingual training and weight sharing, which outperforms dense multilingual models and pruned language-agnostic baselines - with Mu Yang, Andros Tjandra, Chunxi Liu, et al.

  • Improving Fast-slow Encoder based Transducer with Streaming Deliberation (https://lnkd.in/g4VcDX7z): Extension to fast-slow encoder that takes advantage of partial hypotheses to improve WER - with Ke Li, Jay Mahadeokar, JINXI GUO, et al.

  • Spoken Language Understanding Grand Challenge (https://lnkd.in/gbyJKsCY): Aimed at pushing the boundary of on-device end-to-end SLU technology - with Akshat Shrivastava, Paden Tomasello, Suyoun Kim, et al.

Related Questions

What projects has Duc Le worked on at Meta?
What is Duc Le's background in AI and speech recognition?
How did Duc Le contribute to the development of AI Speech org at Meta?
What is Duc Le's role at TikTok Music ML?
What are Duc Le's main research interests?
Duc Le
Duc Le, photo 1
Duc Le, photo 2
Add to my network

Location

San Jose, California, United States