Harry Blum
Founder at Fluxions
Hove, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolHarry Blum is a founder and machine learning engineer with 11 years' experience building audio-first AI systems, from open-source speaker diarization tooling to hyper-realistic voice models deployed at Spotify. He led engineering on emotional and production-grade speech synthesis—work that contributed to a nine-figure acquisition—and later shipped an AI DJ experience to millions. An active open-source contributor and maintainer, he has improved core PyTorch audio libraries and pyannote-audio components for VAD, speaker tracking and verification, emphasizing testability and maintainability. Based in Hove and academically trained in software systems engineering and mathematics, Harry blends rigorous research-quality ML with pragmatic product delivery and a penchant for self-hosting and open-source ecosystems.
11 years of coding experience
5 years of employment as a software developer
Music, Music at The BRIT SchoolBachelor's Degree Mathematics, Bachelor's Degree Mathematics at University of EssexUniversity College London
English
Stackoverflow
3,055reputation
5.9mreached
18answers
5questions
settimeout
top-1%
javascript
top-5%
async-await
top-1%
Github Skills (29)
speaker-diarization10
pytorch10
javascript10
python10
speaker-verification10
machine-learning10
audio-processing10
amazon-dynamodb10
dynamoose10
dynamodb10
settimeout10
async-await10
nodejs10
data-augmentation10
pytorch-lightning9
Programming languages (15)
Github contributions (5)
pyannote/pyannote-audio
Oct 2020 - Apr 2021
Neural building blocks for speaker diarization: speech activity detection, speaker change detection, overlapped speech detection, speaker embedding
Role in this project:
ML Engineer
ML EngineerContributions:32 reviews, 48 commits, 24 PRs in 6 months
Contributions summary:Harry primarily contributed to the project by modifying and adding code related to the core machine learning functionalities of the project. Their work included the addition of documentation and example notebooks, along with the creation of tests for core components. Code changes indicate they worked on the underlying task implementations, including voice activity detection, speaker tracking, and speaker verification. Furthermore, they introduced versioning and refactored the code to enhance code quality and maintainability.
diarizationspeech-recognitionspeaker-verificationspeech-processingspeaker-diarization
asteroid-team/torch-audiomentations
Nov 2020 - Dec 2020
Fast audio data augmentation in PyTorch. Inspired by audiomentations. Useful for deep learning.
Role in this project:ML Engineer
ML EngineerContributions:3 reviews, 18 commits, 3 PRs in 12 days
Contributions summary:Harry primarily contributed to the `torch-audiomentations` library, which focuses on audio data augmentation in PyTorch. Their commits demonstrate a focus on refactoring and improving the existing augmentation transforms, renaming parameters for clarity and consistency across the library. They also implemented batch processing improvements for the `Shift` augmentation, enabling more efficient and correct processing of audio data within the PyTorch framework. The user further contributed by adding sample rate management and providing the library with the capability to handle 2D tensors to process them more efficiently.
pythondifferentiable-data-augmentationaudio-datadspaudio-effects
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.