Heng-jui Chang
Graduate Research Assistant
Cambridge, Massachusetts, United States
Summary
🤩
Rockstar🎓
Top SchoolHeng-jui Chang is a PhD candidate at MIT CSAIL with 11 years of experience building and optimizing speech and machine learning systems. He combines academic research on speech processing with hands-on MLOps and backend engineering, contributing notable improvements to widely used open-source toolkits like s3prl and ESPnet (adding DistilHuBERT/FP16 support and improving ASR recipes and scoring). Multiple research internships at Meta and a teaching assistant role at MIT complement a strong foundation from NTU in electrical engineering. Pragmatic about training efficiency and evaluation, he often focuses on optimizer/scheduler tweaks and language-specific scoring that improve real-world model performance. Colleagues know him for bridging reproducible research and production-ready tooling—and for an unexpected hobby: balloon arts.
11 years of coding experience
4 years of employment as a software developer
Doctor of Philosophy - PhD Computer Science, Doctor of Philosophy - PhD Computer Science at Massachusetts Institute of TechnologyBachelor of Science - BS Electrical Engineering, Bachelor of Science - BS Electrical Engineering at National Taiwan University
Japanese, English, Chinese
Stackoverflow
101reputation
0reached
0answers
0questions
Github Skills (16)
data-preprocessing10
pytorch10
machine-learning10
speech-recognition10
deep-learning10
self-supervised-learning10
automatic-speech-recognition10
python10
asr10
optimization9
bash9
nlp9
distributed-training8
wafv28
kaldi7
Programming languages (4)
Github contributions (5)
s3prl/s3prl
Oct 2021 - Dec 2022
Self-Supervised Speech Pre-training and Representation Learning Toolkit
Role in this project:
Backend Developer & MLOps Engineer
Backend Developer & MLOps EngineerContributions:1 review, 67 commits, 5 PRs in 1 year 1 month
Contributions summary:Heng-jui primarily contributed to the self-supervised speech pre-training toolkit by adding support for DistilHuBERT and FP16 for pretraining. They modified the optimizer configuration and associated scheduling functions, indicating an interest in optimizing the training process. Further contributions involved fixing typos and updating distiller expert, suggesting involvement in model refinement and maintenance.
vq-apcspeech-representationdecoarrepresentation-learningspeech-recognition
espnet/espnet
Jan 2022 - Jan 2022
End-to-End Speech Processing Toolkit
Role in this project:ML Engineer
ML EngineerContributions:3 reviews, 6 commits, 2 PRs in 4 days
Contributions summary:Heng-jui primarily focused on enhancing the SEAME ASR recipe within the ESPnet toolkit. Their contributions include preprocessing script modifications and incorporating new features. The user also made changes to the `run.sh` file, likely configuring and adjusting the training and inference procedures. Furthermore, the commits show an interest in scoring mechanisms for Mandarin and English, suggesting improvements to model evaluation within the ASR pipeline.
speech-recognitionspeech-separationchainerspoken-language-understandingspeech-processing
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.