Sonder: Diffusion-based Song Translation
Sonder is a model that translates singing between languages using a diffusion-based approach, enabling vocal-to-vocal translation while preserving melody and rhythm. Currently under contract with Darkroom Records (Billie Eilish, d4vd, John Summit, etc.)
Endless Lup: Training Creative LLMs
Endless Lup is a 24/7 AI-powered radio station that generates and streams music in the style of Lupe Fiasco. As the project lead, I developed the core audio generation system and trained the language models that power the station. This blog post explores our approach to training large language models for creative tasks, focusing on generating novel and engaging content beyond standard benchmarks. Work covered by Billboard, HipHopDX, and more.
SongGen: Controllable Song Generation (MEng Thesis)
SongGen is an AI-powered songwriting framework that enables interactive, granular song creation through a chat interface, allowing users to guide lyrics, instruments, and genres in real time.
HitgenAI: AI Voice Design Consulting
I run HitgenAI, a consulting firm specializing in AI voice design for musicians, record labels, and companion AI apps. We power new AI identities with custom voice models, enabling creative and commercial applications for artists, producers, and brands. Here are some examples of our work, blending and designing unique AI voices for clients.
AI Billboard Hot 100
Built an AI Billboard Hot 100 at the peak of the AI cover hype. Scraped 10k+ covers from YouTube and TikTok to find the best AI songs on the internet. Shut down by requests from record labels.
