I need speech in the product, not a GPU project.
You want a voice in the app this week. Timbrelay is the hosted call: one project key, one endpoint, and a WAV after watermarking. You do not stand up a runtime for each open model.Product
Hosted open-source speech, for people who ship software.
Timbrelay is for developers who want these models in a product. One project key calls the hosted models. You do not deploy them, and you do not keep a separate client for each one.
The work you skip
Self-hosting is a platform. Calling Timbrelay is a request.
- RuntimeHosted. You send text.
- IntegrationOne endpoint, one project key, one body.
- Model changeChange model. Keep the client.
- ChargeCatalog price before you send. Settle on delivery.
- FileInaudible watermark before it leaves.
What this release is
A clear boundary is part of the product.
- Short text, up to 1,000 characters, returned as a complete WAV. No streaming, batch jobs, or automatic model routing.
- Model availability comes from the enabled catalog. Permanent audio asset storage is outside this release.
- The watermark is a service token. A decode match does not identify a project, person, or single generation.
- No YouTube or TikTok crawler, and no bring-your-own third-party speech key.
- Playground is free for a signed-in project member. Program calls use credits and a project key. The two credentials are not interchangeable.