Kaituo Xu 许开拓
语音算法工程师 at 快手
Haidian District, Beijing, China
Summary
🤩
Rockstar🎓
Top SchoolKaituo Xu is a speech algorithm engineer with 10 years' experience in speech recognition, separation and deep learning, currently working at Kuaishou in Beijing. He has a strong research-to-production track record from internships at Microsoft and Sogou developing TF-LSTM, FSMN and LSTM-based punctuation and acoustic models. Kaituo contributes to open-source speech tools—most notably improving a PyTorch Conv-TasNet implementation with architecture refinements, causal/non-causal training options and bug fixes—demonstrating both model-level and engineering rigor. A Northwestern Polytechnical University alumnus with top-5% undergraduate standing, he blends solid academic foundations with practical optimization and deployment experience in real-world voice products.
10 years of coding experience
1 year of employment as a software developer
Master's degree, Computer Science and Technology, Master's degree, Computer Science and Technology at 西北工业大学
Github Skills (3)
python10
pytorch10
machine-learning9
Programming languages (4)
Github contributions (5)
kaituoxu/Conv-TasNet
Dec 2018 - Dec 2019
A PyTorch implementation of Conv-TasNet described in "TasNet: Surpassing Ideal Time-Frequency Masking for Speech Separation" with Permutation Invariant Training (PIT).
Role in this project:
ML Engineer
ML EngineerContributions:23 commits, 1 PR, 44 pushes in 11 months
Contributions summary:Kaituo primarily contributed to the core code of the Conv-TasNet model. Their work included updating the model's architecture, refining training configurations, and implementing features such as causal and non-causal training options. They also fixed bugs related to the encoder's Conv1d layer and the overlap_and_add function. These changes indicate a focus on improving the model's functionality and performance within the context of speech separation.
pytorchtasnetconvpitdeep-learning
dave-chatmost/TF-punctuator
May 2017 - Nov 2017
Contributions:76 commits in 5 months
predictiondeep-learninglstmmachine-learningtensorflow
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.