Render Audio#

Script or free text to spoken audio — through the platform's Cartesia engine.

POST/v2/audio/render

Request#

Field	Type	Meaning
`touchpoint`	string	Touchpoint ID; the tenant's effective script is rendered
`text`	string	Alternatively: free text (overrides `touchpoint`)
`vars`	object	Placeholder values, e.g. `{"telefon": "+49 30 …"}` — `{{firma}}` and `{{domain}}` come from the tenant branding automatically

bash

curl -s -X POST -H "Content-Type: application/json" -H "X-Tenant: bbe" \
  -d '{"touchpoint": "call.inbound.greeting"}' \
  https://api.callfluence.staging.bestbrandseverywhere.com/v2/audio/render \
  -o greeting.mp3

Response#

The body is the audio itself (audio/mpeg, MP3 mono 128 kbit/s). The X-Engine-Mode header names the path actually used:

`X-Engine-Mode`	Meaning
`direct`	Directly via `api.cartesia.ai` (Sonic-3) — active as soon as a `CARTESIA_API_KEY` is configured
`together-fallback`	Via the verified Together path (`cartesia/sonic-2`)

If the downstream engine fails, the endpoint responds with 502 and a structured error — never with silent empty audio.

Verified round trip#

The render endpoint and voice-to-text are verified against each other: a rendered script transcribes back word for word.