Soniox

Soniox text-to-speech uses the same API key as Soniox speech-to-text. Every voice speaks every supported language. Soniox tuning is passed in synthesizer.options.

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

You can also set options 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

Create an API key in the Soniox console. The key needs text-to-speech access. jambonz sends it as a Bearer token. In the portal, open or add the Soniox speech credential and check Use for text-to-speech; one credential can serve both text-to-speech and speech-to-text.

Field	Required	Meaning
API key (`api_key`)	Yes	Your Soniox key, used for both TTS and STT.
TTS model (`model_id`)	No	The text-to-speech model. Defaults to `tts-rt-v2`.
TTS host (`api_uri`)	No	Host override for text-to-speech, such as `tts-rt.eu.soniox.com`, `tts-rt.jp.soniox.com` or `tts-rt.in.soniox.com` for regional data residency, or a proxy. Leave it out to use `tts-rt.soniox.com`. `ws://` or `http://` selects a plain connection. It does not change the speech-to-text host.

When you save the credential, jambonz checks text-to-speech by synthesizing a short phrase with the Adrian voice.

Models

Model	Notes
`tts-rt-v2`	Default. Realtime model; every voice speaks every supported language.

Voices and languages

A voice is given by name, such as Adrian. A cloned voice from your Soniox account can be used by its id. Every voice speaks every language below, so the portal lists the same voices under each language; Adrian is the default.

Group	Voices
General	`Adrian`, `Maya`, `Daniel`, `Noah`, `Nina`, `Emma`, `Jack`, `Claire`, `Grace`, `Owen`, `Mina`, `Kenji`
Spanish accent	`Rafael`, `Mateo`, `Lucia`, `Sofia`
British	`Oliver`, `Arthur`, `Isla`, `Victoria`
Australian	`Cooper`, `Mason`, `Ruby`, `Elise`
Indian accent	`Arjun`, `Rohan`, `Priya`, `Meera`

Soniox takes a bare language code. jambonz sends the primary subtag of synthesizer.language, so en-US is sent as en and es-ES as es; cmn is sent as zh, nb and nn as no, fil as tl and iw as he. A language Soniox does not support is passed through and Soniox returns an error; jambonz does not fall back to English. Set options.language to send a code as is; it takes precedence over synthesizer.language.

Supported languages: Afrikaans, Albanian, Arabic, Azerbaijani, Basque, Belarusian, Bengali, Bosnian, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, Galician, German, Greek, Gujarati, Hebrew, Hindi, Hungarian, Indonesian, Italian, Japanese, Kannada, Kazakh, Korean, Latvian, Lithuanian, Macedonian, Malay, Malayalam, Marathi, Norwegian, Persian, Polish, Portuguese, Punjabi, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish, Swahili, Swedish, Tagalog, Tamil, Telugu, Thai, Turkish, Ukrainian, Urdu, Vietnamese and Welsh.

Options

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

model_id

string

Text-to-speech model. Overrides the credential’s model.

language

string

Language code sent to Soniox as is, such as fr. Overrides the code derived from synthesizer.language.

speed

number

Speaking rate, 0.7 – 1.3. Soniox’s default is 1.0.

reduce_silence

boolean

Shortens pauses in the generated speech.

client_reference_id

string

Your own reference, up to 256 characters, attached to the request in Soniox’s usage records.

{
  "speed": 1.2,
  "reduce_silence": true
}

Streaming

Soniox supports TTS streaming, so it can be driven token by token from an LLM using tts:tokens, for example from the agent verb. jambonz streams text to Soniox over a live connection, starting a new Soniox stream for each turn, and on barge-in sends Soniox’s cancel instead of reconnecting.

Soniox returns character-level timestamps, so alignment events are available when alignment is enabled.

Soniox limits a single stream to two minutes of audio; longer speech is cut off with an error. A stream that receives no text for several seconds is closed by Soniox, and the next text opens a new one.