Presenting LatentScore at NIME 2026 in London

I’ll be presenting LatentScore at NIME 2026 in London this June. Read more about how it works or try it out.

For those unfamiliar, LatentScore lets you generate procedural music from text, all on CPU. No audio samples, no GPU. You can even use language models to drive it.

The data behind it was built in a kind of unusual way - I scraped textbooks and taught LLMs to generate structured configurations for an artificial musical synthesizer. Books cover such a wide slice of human experience that they turned out to be a surprisingly good source for extracting a ton of musical “vibes” without ever touching an audio dataset.

If you’re curious or have questions, feel free to reach out at prabal@rjeinc.ca.