Desh Raj
Senior Research Scientist at NVIDIA
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolDesh Raj is a Senior Research Scientist with 11 years of experience building state-of-the-art speech and full-duplex voice models, currently working on NeMo Speech at NVIDIA after leading post-training and evaluation efforts for LLaMA-based voice systems at Meta. He combines deep academic grounding (Ph.D. in CS from Johns Hopkins) with hands-on engineering contributions to prominent open-source speech toolkits like Kaldi, Lhotse, and Icefall, including performance optimizations and end-to-end recipe implementations. His work spans model fine-tuning, rejection sampling at scale, tool-calling from voice, and practical on-device compression techniques, reflecting both research depth and production sensibility. Based in New York, he has a track record of shipping pretrained models to Hugging Face and improving decoding speed and evaluation pipelines, demonstrating a knack for squeezing performance and usability gains from complex speech stacks.
11 years of coding experience
3 years of employment as a software developer
Bachelor of Technology (B.Tech.) Computer Science and Engineering, Bachelor of Technology (B.Tech.) Computer Science and Engineering at Indian Institute of Technology, GuwahatiJohns Hopkins University
Github Skills (25)
algorithm10
algorithms10
pytorch10
c-language10
python10
machine-learning10
automatic-speech-recognition10
data-structure10
audio10
kaldi10
deeplearning-ai10
deep-learning10
speech10
ai10
asr10
Programming languages (12)
Github contributions (5)
lhotse-speech/lhotse
Jan 2021 - Jan 2023
Tools for handling speech data in machine learning projects.
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:201 reviews, 293 commits, 130 PRs in 2 years
Contributions summary:Desh's commits indicate they are primarily responsible for extending and modifying the AMI recipe, a core component of the project. They are focused on audio processing and speech recognition, as suggested by the codebase's purpose and the inclusion of deep learning and speech-related libraries. The contributions also involve modifications to the underlying infrastructure to accommodate the new functionality, as illustrated by the incorporation of microphone settings and official scenarios.
pytorchkaldipythondatamachine-learning-projects
k2-fsa/icefall
May 2022 - Dec 2022
Role in this project:ML Engineer
ML EngineerContributions:51 reviews, 13 commits, 15 PRs in 6 months
Contributions summary:Desh's contributions center around implementing a SPGISpeech recipe within the `icefall` repository, indicating work in speech recognition. This involves initializing the recipe, adding decoding functionalities, integrating a SPGISpeech transducer model, and incorporating results and tensorboard visualizations. The user also added a pretrained model to Hugging Face and included CER numbers for evaluation.
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.