Groq

Groq serves Canopy Labs’ Orpheus text-to-speech models through its OpenAI-compatible speech endpoint (POST /openai/v1/audio/speech). Groq tuning is passed in synthesizer.options.

{
  "verb": "say",
  "text": "Hi there, your order ships today.",
  "synthesizer": {
    "vendor": "groq",
    "language": "en-US",
    "voice": "autumn",
    "options": {
      "speed": 1.1
    }
  }
}

You can also set these in the portal under Extra Options on the speech credential, in which case they apply to every request that uses that credential. Settings supplied on the verb take precedence.

Credential

Add a Groq speech credential in the portal or API, using the vendor id groq. Groq is used for text-to-speech only.

Field	Required	Meaning
API key (`api_key`)	Yes	Your Groq key (`gsk_...`), sent as `Authorization: Bearer <api key>`.
API base URL (`api_uri`)	No	Host override, such as a proxy. `http://` is used when given, otherwise `https://`. Leave it out to use `api.groq.com`.
Model (`model_id`)	No	Default model for this credential. Leave it out to pick the model by language.

When you save the credential, jambonz checks it by synthesizing a short phrase in en-US with the autumn voice. A credential whose model is canopylabs/orpheus-arabic-saudi is checked in ar-SA with the fahad voice instead.

A Groq speech credential is separate from a Groq LLM service used by the agent verb. Add each one where it is used, even when they share the same API key.

Groq requires you to accept each Orpheus model’s terms in the Groq console before first use. Until then, requests for that model fail.

Voices and languages

Language	Model	Voices
English (US) `en-US`	`canopylabs/orpheus-v1-english`	`autumn` (default), `diana`, `hannah`, `austin`, `daniel`, `troy`
Arabic (Saudi) `ar-SA`	`canopylabs/orpheus-arabic-saudi`	`fahad` (default), `sultan`, `abdullah`, `lulwa`, `noura`, `aisha`

jambonz does not send synthesizer.language to Groq; it uses it only to pick the model. Any Arabic language (ar, ar-SA, …) uses canopylabs/orpheus-arabic-saudi, and every other language uses canopylabs/orpheus-v1-english. Set model_id to choose the model yourself.

Orpheus accepts bracketed vocal directions in the text, such as [cheerful], and jambonz passes them through unchanged.

jambonz requests WAV audio, removes the header and resamples the audio to the call’s rate. Without sample_rate, Groq’s default of 48 kHz is assumed.

Options

Every option is optional. An option you don’t set is not sent, so Groq’s own default applies.

model_id

string

canopylabs/orpheus-v1-english or canopylabs/orpheus-arabic-saudi. Overrides the credential’s model and the choice by language.

speed

number

Speaking rate, from 0.5 to 5. 1 is normal speed.

sample_rate

integer

Output sample rate: 8000, 16000, 22050, 24000, 32000, 44100 or 48000.

{
  "model_id": "canopylabs/orpheus-arabic-saudi",
  "speed": 0.9,
  "sample_rate": 24000
}

Streaming

Groq has no token-streaming websocket, so it does not support TTS streaming: use it for one-shot say, not for say with stream: true or the agent verb.