Daniel Povey
Chief Speech Scientist at Xiaomi Technology
Beijing, China
Summary
🤩
Rockstar🎓
Top SchoolDaniel Povey is a chief speech scientist based in Beijing with 15 years of experience advancing automatic speech recognition from academic labs to industry. He holds a PhD in speech recognition from Cambridge and has a long research pedigree at IBM, Microsoft Research, and Johns Hopkins, where he contributed extensively to the open-source Kaldi toolkit. At Xiaomi he leads applied speech research and model engineering, blending deep expertise in TDNN-F and SpecAugment-style techniques with hands-on code changes and performance tuning. His GitHub contributions show practical work on high-profile projects like Kaldi and the icefall/k2-fsa ecosystem, indicating comfort across model internals, feature extraction, and training pipelines. Known for turning research ideas into production-ready systems, he combines rigorous scientific methods with pragmatic engineering to ship reliable ASR components at scale. An often-overlooked strength is his sustained commitment to open-source tooling, which has helped shape reproducible speech-research workflows used across the community.
15 years of coding experience
18 years of employment as a software developer
PhD, Speech Recognition, PhD, Speech Recognition at University of CambridgeBachelor of Arts (BA), Natural Sciences Tripos, Bachelor of Arts (BA), Natural Sciences Tripos at Cambridge University
Github Skills (17)
pytorch10
python10
scripting10
attention-mechanism10
machine-learning10
rnn-model10
n10
mask-rcnn10
script10
conformer10
sh10
faster-rcnn10
speech-recognition10
shell10
data-augmentation10
Programming languages (9)
Github contributions (5)
kaldi-asr/kaldi
Aug 2020 - Aug 2021
kaldi-asr/kaldi is the official location of the Kaldi project.
Role in this project:
ML Engineer
ML EngineerContributions:68 reviews, 9 commits, 2854 PRs in 1 year 1 month
Contributions summary:Daniel's commits indicate a focus on implementing and modifying machine-learning related components within the Kaldi project, especially those related to speech recognition. Their contributions include fixing bugs within dropout-related code and modifying and augmenting data preparation and model training scripts, with a focus on the TDNN-F based models. The user demonstrated skills in applying and refining techniques relating to speech recognition and deep learning.
cudakaldiasrspeech-to-textkaldi-asr
k2-fsa/icefall
Jul 2021 - Nov 2022
Role in this project:ML Engineer
ML EngineerContributions:119 reviews, 221 commits, 37 PRs in 1 year 4 months
Contributions summary:Daniel primarily focused on modifying the fbank computation by integrating a chunky writer and then made a variety of changes to the model, including adding and removing functions for attention. Further modifications included modifications to the SpecAugment setup and layer normalization. The user appears to be involved in tuning and iterating upon the existing models.
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.