Building a B2B API for audio + text-to-timbre search — would this be useful in your plugin or sample platform?

Explore how Machine Learning and AI can expand musical creativity while keeping the human in the creative workflow. This forum is dedicated to respectful dialogue where diverse perspectives are welcomed.
Post Reply New Topic
RELATED
PRODUCTS

Post

Hi everyone,

I'm Yannick, co-founder of Impulsia (impulsia.ai). We've been quietly working on something for a while and we'd rather get honest feedback from this community now than ship into the void in six months.

The problem I keep hearing

Producers, sound designers and even pros tell me they spend 20–30% of their session time scrolling through preset banks and sample libraries. The current search experience is mostly tag-based (genre, mood, instrument) — which barely captures what people actually want, which is *timbre*. "Bright analog pad", "gritty Rhodes", "airy female chop" — none of those map cleanly to existing taxonomies.

What we built

A B2B API (codename: Granulator) that ingests any audio collection — samples, presets, loops, stems — and lets you query it two ways:

- Audio-to-timbre: upload a sound, get the closest timbral matches
- Mix-to-timbre: upload a mix of instruments, get the closest timbral matches for each instrument
- Text-to-timbre: natural language queries like "swelling cinema brass pad" or "lofi tape-saturated Wurlitzer"

Under the hood are deep neural audio models, trained on synth/sample audio data rather than generic music. The idea is that a plugin maker, a sample marketplace, or a preset-bank author could plug this in and offer real timbre-aware search inside their own product, without having to build the ML themselves.

What I genuinely want to know from you

1. If you're a plugin developer or run a sample/preset platform: would this solve a real problem, or are you already happy with what you've built?
2. What would make you actually integrate something like this — pricing model, on-prem option, custom-trained embeddings on your own catalog, latency requirements?
3. What would be a deal-breaker — data ownership, embedding portability, vendor lock-in?
4. Have you seen similar tools in the wild? I'd love to understand who else is in this space.

No pitch, no demo link being pushed — happy to share more privately if anyone's curious, but mostly I just want the unfiltered take from the people who'd actually use this. Reply here or DM me, whichever's easier.

Thanks,
— Yannick
impulsia.ai

Post Reply

Return to “Machine Learning and AI for Music Creation”