Palabra × jambonz

A jambonz custom speech vendor backed by Palabra Realtime STT/TTS: plug Palabra into the jambonz CPaaS as the recognizer and synthesizer of your applications.

• /stt (WebSocket, no subprotocol) — jambonz sends start (language, format: "raw", encoding: "LINEAR16", sampleRateHz, interimResults), binary linear16 frames, then stop. Audio is forwarded to Palabra at the call's sample rate (8 kHz telephony as a rule — no resampling); Palabra transcripts come back as transcription messages (interim ones only when interimResults is on). On stop the bridge sends finalize, delivers the final transcript and closes the socket, as the protocol requires.
• /tts (HTTP POST) — jambonz sends {language, voice, type: "text"|"ssml", text}; the bridge returns the whole utterance as audio/wav (SSML tags are stripped — Palabra TTS takes plain text).
• /healthz (GET).

Run

pip install palabra-ai-jambonz-bridge
export PALABRA_API_KEY=plbr_...
export JAMBONZ_API_KEY=...          # the API key you enter for the vendor in the jambonz portal
palabra-jambonz-bridge --port 8081

docker run -e PALABRA_API_KEY=plbr_... -e JAMBONZ_API_KEY=... -p 8081:8081 palabra-jambonz-bridge

jambonz configuration

Portal: Speech → Add speech service → vendor "Custom" → name, API key (the same value as JAMBONZ_API_KEY), STT URL wss://<your-host>/stt, TTS URL https://<your-host>/tts. Applications then use custom:<name> as the recognizer / synthesizer vendor.

Protocol details

Transcription messages:

{ "type": "transcription", "is_final": true,
  "alternatives": [{ "transcript": "...", "confidence": 0.9 }],
  "channel": 1, "language": "en-US" }

language echoes the BCP-47 code jambonz sent; Palabra receives the short code (en-US → en). Palabra does not report confidence, so a constant 0.9 is returned. Errors are {"type": "error", "error": "..."}. options.hints are accepted and ignored. jambonz's separate streaming-TTS WebSocket protocol (connect / stream / flush / stop) is not implemented.

Tests

pip install -e ".[dev]"
pytest -m "not live"                          # offline (fake Palabra client, real FastAPI app)
PALABRA_API_KEY=plbr_... pytest -m live -s    # /tts (8 kHz wav) -> /stt round trip against the API

Status

Protocol checked against the current jambonz documentation and the reference custom-speech-example; exercised end-to-end against the Palabra API with a test client playing jambonz's part (8 kHz). Not yet run against a live jambonz. Concurrent calls each open their own Palabra STT session with the same API Key.