For developers

Open-source TTS. One hosted API.

Timbrelay turns multiple open-source speech models into one callable hosted API, so you skip local deployment and a new client for every model.

Open Playground Switch a model

No local deploy
One request shape
Price before you send
Watermark on delivery

Request previewNot a live call

“Welcome. It is good to have you here.”

endpoint: /v1/audio/speech
model: cosyvoice3
voice: voice_preset_en_5639
format: wav

The key and body stay. A preset must be enabled for the model you pick.

The job to be done

“I want these models. I do not want to deploy them, or write a new integration for each one.”

01Skip the deployHosted open-source speech, scoped to a project key.02Switch the modelFour connected models. One request shape.03Mark every fileInaudible watermarking finishes before delivery.04See the chargePrice first. Settlement only on a delivered file.

Built into the file

Every delivered file already carries the mark.

Watermarking is not a step you add later. Encoding finishes before the WAV is returned. If it cannot be confirmed, the model audio does not leave, and a paid hold is released.

See the delivery rule

IndexTTS-2.5English reference · watermarked sample

Welcome to Timbrelay. Give your words a voice of their own.

Hear the marked file0:04

Watermark applied before this sample was published.

Start with a listen

Try a voice. Keep the same call when you ship it.

Open Playground Read the request