Crowd chants for 10,000+ nicknames
A proof of concept for a 3D crowd system: ElevenLabs-generated chants covering more than 10,000 player nickname combinations, a scale that would be impractical to record with live actors.
AI Audio
HUMAN-LED · AI-ENHANCED
Human-led, AI-enhanced. I’m the human in the loop: I reach for generative tools when a job would be impractical to record or write by hand, then direct, review and shape what they produce. I don’t deliver anything made solely by AI.
It started inside game projects, solving problems live recording couldn’t. Now it’s commissioned work in its own right.
Where it started
Inside game projectsPrompt-to-MIDI composition assistant
An assistant I built that turns music prompts and reference data into four-track MIDI setups: bass roots, voiced chords and melody. On the Jetpack Joyride Racing score it let me try five to ten variations to find the right foundation for a track.
Coach dialogue validation
Used LLMs to shape the Coach's voice and to validate the concatenated dialogue system, so thousands of possible phrase combinations still read naturally and stay in character.
Commissioned work
Client briefsCommissioned music catalogs
Catalogs written to client briefs. I built the 12-stage pipeline that writes the lyrics and style prompts behind automated quality gates, and every reviewer correction is logged as a corpus for tuning the prompts.
- 90-track retail brief
- 2 × 100-track fitness briefs
Client audio-drama series
I built the production pipeline for a client's audio-drama series: ElevenLabs v3 voices and generated sound effects, assembled directly into REAPER projects.
Raw vs. finished
A/B listeningComing soon Switch between the raw AI take and my finished version at the same point in the track.