TTS Vendor Settings

Most TTS vendors expose knobs that go beyond picking a voice — speaking rate, expressiveness, stability, and so on. jambonz passes these through untouched in synthesizer.options.

{
  "verb": "say",
  "text": "Thanks for calling, how can I help you today?",
  "synthesizer": {
    "vendor": "cartesia",
    "language": "en",
    "voice": "a0e99841-438c-4a64-b679-ae501e7d6091",
    "options": {
      "speed": -0.3,
      "emotion": "positivity:high"
    }
  }
}

The fields options accepts are defined entirely by the vendor — jambonz does not validate them. See each vendor’s page below for what it supports.

Where settings can be set

There are two places, and they combine:

On the speech credential, under Extra Options in the portal (Account → Speech → your credential). These apply to every request using that credential.
On the verb, in synthesizer.options. These apply to that verb only.

Where both are present, the verb wins.

Setting them on the credential is the better default when you want one consistent voice character across an application. Use the verb when a particular prompt needs to differ — a slower, clearer reading of an account number, for example.

Options are passed to the vendor as-is. If you send a field a vendor doesn’t recognise, most vendors ignore it silently rather than returning an error — so a typo usually shows up as “my setting had no effect”, not as a failure.

Vendors with settings

AssemblyAI

Library and cloned voices, EU region, session timeout.

Async

Flash and Pro models, fifteen languages, speed and stability.

Azure OpenAI

OpenAI voices on your own Azure deployment, speed, spoken instructions.

Bland AI

Voice UUIDs, expressiveness and stability controls.

Camb.ai

MARS voices, speaking rate, enhancement, inference tuning.

Cartesia

Speaking rate and emotion, with model-family differences.

ElevenLabs

Stability, similarity boost, style, speaker boost, latency optimisation.

Fish Audio

Cloned voices, latency mode, speed and volume.

Giggy

Speaking rate and a seed for repeatable output.

Gnani

Indic and Indian English Timbre voices, speed, language.

Google

Gemini TTS model and prompt.

Gradium

Temperature, voice similarity, speaking rate, pronunciation dictionaries.

Groq

Orpheus voices for English and Saudi Arabic, speed, sample rate.

Hume

Octave voices, acting directions, speed, trailing silence.

Inworld

Pitch, speaking rate, temperature.

Kokoro

Self-hosted, OpenAI-compatible; speed, volume, voice mixing, text normalization.

KugelAudio

Speed, guidance scale, temperature, normalization, pronunciation dictionaries.

MiniMax

Multilingual voices, emotion, speed, volume, pitch, voice effects, pronunciation dictionaries.

Mistral

Voxtral preset and cloned voices, expressiveness, emotion tags, pronunciation hints.

Murf

Style, rate, pitch, variation.

Neuphonic

Multilingual voices, speed, temperature, IPA dialect.

Paxa Labs

Thai, English and Mandarin voices, direction tags, speed, reading language.

Resemble

Project selection and HD synthesis.

Respeecher

Real-time models per language, sampling controls.

Rime Labs

Speed, latency reduction, and per-model sampling controls.

Sarvam AI

Indic and Indian English voices, pace, pitch, loudness, temperature, pronunciation dictionaries.

Simplismart

Orpheus, Veena and Qwen 3 TTS models, sampling controls, dedicated deployments.

Smallest AI

Lightning voices, speed, number and date reading, pronunciation dictionaries, content filter.

Soniox

Voices in sixty languages, speed, silence reduction, regional hosts, same key as Soniox STT.

Speechify

Loudness and text normalization, model selection.

Speechmatics

British and American English preview voices, same key as Speechmatics STT.

Spitch

English and African-language voices, speed.

Svara

Speed, pronunciation dictionaries, eager streaming, sampling controls.

Vakyam

Indian English and seven Indic languages, speed.

xAI

Speed, latency optimisation, text normalization.

Vendors not listed here take no options beyond voice, language and model.

Fallback vendors

synthesizer.options applies to the primary vendor only. If you configure a fallbackVendor, it is used with its own credential-level Extra Options — the primary vendor’s options are not carried over, since the two vendors’ options are unrelated.