takt labs

Takt Labs

Experiments in steering AI music.

Small tools for prompting AI music models in new ways — built at the intersection of music production and machine learning. Two so far; more when they're ready.

/ tools

01 audio prompting

AudioPrompt

Steer AI music generations toward a key, tempo, and groove.

AudioPrompt generates short clips of shaped noise — with melody, tempo, and groove imprinted into them — designed to be used as audio prompts. Feed one into an AI music model and generate over it to push your result toward a particular musical direction. Unlike a reference track, these are built specifically to carry pitch and rhythm information the model can follow.

Python · NumPy · SciPy · STFT spectral imprinting on pink noise

Try it →Suno playlist ↗Source ↗
Dark magma spectrogram output from AudioPrompt, showing harmonic bands imprinted into shaped pink noise.

02 prompt vocabulary

Genre Vector Explorer

Discover related style tags to sharpen your text prompts.

Type a few descriptors — “jazz, smooth, mellow” — and it surfaces the style words most closely related to them, drawn from a model trained on Wikipedia's music-genre writing. A fast way to find the right vocabulary for a text prompt, or just to explore how genres and moods connect.

Word2Vec · cosine similarity · trained on Wikipedia music articles

Try it →Source ↗
Genre Vector Explorer neighborhood map showing related genre and mood terms as labeled points in a semantic space.

/ contact

Open to collaboration, work, or just to talk about music and AI.