PRODUCT
Voices real enough to need consent: inside Blab, our voice studio
1 min read
Blab is the studio surface for our own text-to-speech family — sub-second streaming, consented voice cloning, and dubbing that keeps the original soundtrack.
Blab is two things with one name: a model family we train, and the studio you use to drive it. The studio is the product; the Blab text-to-speech family is the engine underneath.
Built on a model we own
Most voice products resell someone else’s synthesis. Blab is the studio surface for our own Blab TTS family — one of the five in-house model families under the platform. Owning the model is why we can stream speech in under a second and keep improving the voices instead of waiting on a vendor.
What you can do with it
Blab covers the real voice workflows:
- Stream text-to-speech in under a second for live, conversational use
- Clone voices — with consent, because the voices are convincing enough to require it
- Dub dialogue across languages while preserving the original soundtrack
- Run 40+ languages and sixty-four voices, voice agents and batch jobs
Wired for production
It is a studio with an API behind it: a REST API, a TypeScript SDK, MCP support and signed webhooks, so the same voices that power a one-off project can power an application or another Arpanet product. Calleague’s phone agents, for instance, speak through the same family.
Consent is the point, not the footnote
When synthesis gets this good, consent stops being a checkbox. Blab treats voice cloning as something you opt into deliberately — a design choice that matters more as the technology improves, and one that fits a platform engineered for the KVKK and the GDPR from the start.
Voices so real, we require consent.
Blab is what owning the model layer looks like at the surface: a studio that gets better because the engine underneath is ours.