Hey all,
Polling the JUCE community before we commit to building something. At Impulsia (impulsia.ai) we’ve been developing AI models for timbre-aware audio + text search, and we’re trying to decide whether to wrap them in a public B2B API aimed at plugin and sample-library developers. Before we build out the API surface and any C++ tooling, we’d like to know if there’s real demand here — and what shape it would need to take.
What the underlying tech does (today, internally)
We have working models that index any audio collection — presets, samples, loops, stems — and support three query modes:
- Audio-to-timbre — send audio sample, get nearest timbral matches
- Mix-to-timbre — send a multi-instruments mix, get per-source matches (internal demixing)
- Text-to-timbre — natural language like “gritty lofi Rhodes” or “swelling cinema brass pad”
Models are trained on synth/sample data rather than generic music, so the embedding space is closer to preset retrieval than to MIR. We’ve built demos and benchmarks; we have not built a public API or SDK — that’s the open question.
What we’re considering building
A B2B API, plus eventually a JUCE-friendly C++ SDK, so plugins or apps could add things like:
- “Find similar preset” in a synth or sampler browser
- Smart search bar in a sample-pack plugin
- Recommendations / “presets like this one”
- Custom embeddings trained on your own catalog
Cloud REST first; on-prem and on-device down the road if it makes sense.
What we’d like to know from you
If you’re shipping or planning a JUCE plugin or sample/preset product — and would (or wouldn’t) want to plug in something like this:
- Is preset/sample search actually a problem you’d pay to solve, or have you got it covered?
- Cloud REST vs. on-prem vs. edge-bundled model — what would you realistically ship?
- C++ SDK vs. raw REST — what would you reach for?
- Threading / latency / offline-mode constraints we should know about?
- Pricing models that would or wouldn’t fly (per-call, flat tier, per-end-user, hosted-catalog)?
- Anything similar already in the wild we should look at?
No demo link, no pitch — we’re genuinely scouting demand before committing engineering. If you’d consider being an early design partner once we do build it, even better. Reply here, DM, or yannick.molle@impulsia.ai.
Thanks,
— Yannick
impulsia.ai
