Amir Hussein
Baltimore, Maryland, United States
Summary
🤩
Rockstar🎓
Top SchoolAmir Hussein is a Ph.D. candidate at Johns Hopkins CLSP with seven years of experience building real-time, context-aware speech translation and low-resource ASR systems for conversational and long-form audio. His work spans end-to-end speech translation, multilingual and code-switched ASR, and practical data engineering—he contributed MGB2 dataset integration to the popular Lhotse toolkit and improved data pipelines in ESPnet. He has a track record of bridging research and product: from pioneering a transformer-based Arabic ASR with novel VAD at Kanari AI to deploying bridged back-translation and context-aware E2E-ST methods in his academic projects. Internships at NVIDIA and MERL reflect a blend of industry-scale streaming ST and multimodal representation research. Comfortable with both model training and the messy realities of data prep, he leverages creative augmentation and system combination techniques to tackle low-resource and dialectal challenges.
7 years of coding experience
4 years of employment as a software developer
B.S (Honour), Electrical and Electronic Engineering, Discipline Control Engineering and Instrumentation, B.S (Honour), Electrical and Electronic Engineering, Discipline Control Engineering and Instrumentation at University of KhartoumHigh School, Student, High School, Student at Yousif Al DigerMaster's degree, Computer Engineering, Master's degree, Computer Engineering at American University of Beirut
English, Ukrainian, Arabic, Ukrainian
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (26)
pytorch10
python10
scripting10
data-engineering10
machine-learning10
beautifulsoup10
data-preprocessing10
script10
dataprep10
speech10
data-processing10
sh10
speech-recognition10
shell10
bash9
Programming languages (6)
Github contributions (5)
espnet/espnet
Aug 2020 - Feb 2022
End-to-End Speech Processing Toolkit
Role in this project:
ML Engineer
ML EngineerContributions:6 reviews, 61 commits, 2 PRs in 1 year 6 months
Contributions summary:Amir primarily contributed to the data preparation and training pipeline for the end-to-end speech processing toolkit. Their work included modifying the data preparation scripts (mgb_data_prep.sh), updating the run scripts (run.sh) for training, and adjusting the text segmentation process. The user also updated configuration files and implemented speed perturbation for data augmentation. These changes directly impacted the data ingestion, model training and potentially model performance.
speech-recognitionspeech-separationchainerspoken-language-understandingspeech-processing
lhotse-speech/lhotse
May 2022 - Jun 2022
Tools for handling speech data in machine learning projects.
Role in this project:Data Engineer
Data EngineerContributions:40 reviews, 32 commits, 4 PRs in 18 days
Contributions summary:Amir primarily contributed to preparing and integrating the MGB2 dataset within the Lhotse speech data handling framework. Their work involved creating a recipe to download, preprocess, and create manifests for the dataset, including handling text cleaning and BuckWalter transliteration. They also addressed issues such as fixing unicode strings and ensuring the correct data loading using `load_kaldi_data_dir()` and `copy()`. The contributions enabled utilizing MGB2 data with the Lhotse library for speech processing tasks.
pytorchkaldipythondatamachine-learning-projects
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.