Chenda Liao
Principal Research Scientist - Speech Architect at Zoom
Shanghai, United States
Summary
🤩
Rockstar🎓
Top SchoolChenda Liao is a Principal Research Scientist and Speech Architect with 11 years of experience designing production-grade speech AI across leading cloud and consumer platforms. She has driven speech innovation at Zoom and Microsoft Azure AI, and previously led intelligent speech interaction research at Alibaba DAMO and Nuance, blending deep research with hands-on engineering. Her PhD-level training in control systems and automation informs rigorous system design, particularly for multi-speaker and mixed-speech scenarios. An active contributor to the widely used ESPnet toolkit, she implemented a Transformer-based multi-speaker ASR solution with PIT-CTC and multi-task learning, bridging research ideas into open-source code. Based in Shanghai with international experience, she excels at turning cutting-edge speech models into scalable, deployable services.
11 years of coding experience
3 years of employment as a software developer
Master study in Control Theory and Control Engineering, Master study in Control Theory and Control Engineering at Shanghai Jiao Tong UniversityBachelor of Engineering Automation, Bachelor of Engineering Automation at Zhejiang UniversityDoctor of Philosophy (PhD) Mechanical Engineering(system and control), Doctor of Philosophy (PhD) Mechanical Engineering(system and control) at University of Florida
Chinese, English
Github Skills (10)
transformers10
transformer10
machine-learning10
voice-recognition10
speech-recognition10
pytorch10
deep-learning10
automatic-speech-recognition10
asr10
speech-processing9
Programming languages (6)
Github contributions (5)
espnet/espnet
Jun 2020 - Aug 2022
End-to-End Speech Processing Toolkit
Role in this project:
ML Engineer
ML EngineerContributions:133 reviews, 217 commits, 27 PRs in 2 years 2 months
Contributions summary:Chenda contributed to the development of a Transformer-based speech recognition model for single-channel multi-speaker mixture speech within the espnet framework. Their work involved implementing a fusion of existing modules, including the Transformer-based Encoder with three stages for encoding, separating, and transforming mixed speech. They also incorporated the use of Permutation Invariant Training (PIT) in CTC to determine optimal permutations and included the implementation of a multi-task learning approach with CTC and attention-based decoder. The modifications are reflected in the updated `espnet/nets/pytorch_backend/e2e_asr_mix_transformer.py` file.
speech-recognitionspeech-separationchainerspoken-language-understandingspeech-processing
LiChenda/notebook
Jun 2021 - Jun 2023
Contributions:5 pushes in 2 years
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.