Voice providers — bring your own STT/TTS

Let Chamade run hosted STT (transcripts) and hosted TTS (chamade_call_say) on your meetings using your own API keys. Configure once in the dashboard; every new call picks it up automatically.

What voice providers do

A voice gateway like Chamade bridges a meeting's audio to your agent. On each call you can either:

• BYO-audio — you receive raw PCM on the call's audio WebSocket and run your own STT/TTS locally (OpenAI Realtime, LiveKit, Pipecat, whatever).
• Hosted STT/TTS — you add an API key for a provider once in the dashboard. Chamade handles transcription and speech synthesis on its side using your key: transcripts arrive as call_transcript events, chamade_call_say speaks text in the meeting.

The two modes are independent per side: hosted STT + BYO TTS is fine, and vice versa. No preset for a given side = that side stays BYO-audio on the WebSocket.

Add a preset

Go to your dashboard → Voice providers (STT / TTS), click + Add on the STT or TTS column:

1. Pick a provider.
2. Paste your API key. Chamade stores it Fernet-encrypted and never returns it over the API.
3. Click Save. Chamade probes the key against the provider's auth endpoint; bad keys are rejected immediately.

Every other option (model, voice, language, stability, …) has a sensible default — you only need to open Advanced options if you want to customize. For TTS, use the Load voices from provider button to pick a specific voice ID from the provider's catalog.

Enabled = the one being used

At most one preset per kind (STT, TTS) can be enabled at a time. Enabling a new preset disables the previous automatically. Disabled presets keep their data (key, options) — useful to quickly swap between a "cheap" Deepgram and a "premium" ElevenLabs without re-entering keys.

No enabled preset for a given kind = that side stays BYO-audio on the next call.

Capabilities on each call

When you create a call with POST /api/call (or chamade_call_join), the response capabilities array advertises hosted_stt / hosted_tts if a preset is enabled for that kind. Your agent can branch on that:

"capabilities": ["audio_in", "audio_out", "read", "write", "typing", "hosted_stt", "hosted_tts"]

• hosted_stt present → call_transcript events will arrive on the WS (and in chamade_call_status).
• hosted_tts present → chamade_call_say works; otherwise it returns 400 and you should synthesize locally + send PCM on the audio WS.

Per-call override

If you keep multiple presets around (say, a premium and a cheap TTS) and you want to force a specific one for one call, pass the preset UUID at creation time:

POST /api/call
{
  "platform": "teams",
  "meeting_url": "...",
  "tts_voice_config_id": "<uuid-of-a-TTS-preset>",
  "stt_voice_config_id": "<uuid-of-an-STT-preset>"
}

The UUID comes from GET /api/voice-configs. Trying to use a disabled preset returns 400 — enable it in the dashboard first.

Supported providers

More providers (Azure, Google, Speechmatics, Gladia, …) can be added on request — drop a note to [email protected] if you need a specific one.

REST API

The dashboard is a thin wrapper over these endpoints — call them directly from your own tooling if needed: