Camb.ai

Camb.ai generates speech with its MARS models in more than thirty locales, including English, Spanish, French, German, Arabic, Hindi and other Indian languages, Japanese, Korean and Mandarin Chinese. Camb.ai tuning is passed in synthesizer.options.

{
  "verb": "say",
  "text": "Hi there, your order ships today.",
  "synthesizer": {
    "vendor": "camb",
    "language": "en-US",
    "voice": "147320",
    "options": {
      "speaking_rate": 1.0
    }
  }
}

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

Create an API key in the Camb.ai studio. jambonz sends the key in the x-api-key header. Add it as a Camb.ai speech credential in the portal or API. Camb.ai is used for text-to-speech only.

Models

Model	Notes
`mars-flash`	Default. Low latency, 22.05 kHz.
`mars-pro`	Higher quality, 48 kHz.
`mars-instruct`	Reads `user_instructions` to direct the delivery, 22.05 kHz.
`mars-8.1-flash-beta`	Beta, 48 kHz.
`mars-8.1-pro-beta`	Beta, 48 kHz.

Pick the model on the speech credential, or set options.model_id. The model applies to say without streaming; streaming always uses mars-8.1-flash-beta. jambonz resamples audio to the call’s rate.

Voices and languages

A voice is a numeric Camb.ai voice id. The portal lists 147320, Camb.ai’s default voice, for every locale; MARS voices are cross-lingual, so the same voice speaks whichever language is set. Voices you have cloned or created in your Camb.ai account work when you enter their id as a custom voice.

synthesizer.language is sent lowercased, so en-US is sent as en-us and pt-BR as pt-br. Set options.language to send a Camb.ai language code directly; it takes precedence over synthesizer.language. If neither is set, en-us is used. A locale Camb.ai does not support is rejected by Camb.ai.

Options

language

string

Camb.ai language code, sent as written, such as en-us or fr-ca. Overrides synthesizer.language.

model_id

string

mars-flash, mars-pro, mars-instruct, mars-8.1-flash-beta or mars-8.1-pro-beta. Overrides the model on the credential. Not used when streaming.

speaking_rate

number

Speaking rate relative to the voice’s own speed.

enhance_named_entities_pronunciation

boolean

Improve the pronunciation of names and other named entities.

apply_enhancement

boolean

Apply Camb.ai’s audio enhancement to the output.

enhance_reference_audio_quality

boolean

Enhance the quality of the voice’s reference audio before synthesis.

maintain_source_accent

boolean

Keep the voice’s original accent when it speaks another language.

inference_steps

integer

Number of inference steps, 5 – 200. More steps trade latency for quality.

idle_timeout

number

Seconds Camb.ai waits for more text before it synthesizes what it has. Streaming only. Every turn’s first audio is delayed by this much, so keep it small (for example 0.3 – 1); a large value such as 30 delays speech by 30 seconds.

word_timestamps

boolean

Return word-level timestamps. Streaming only. When unset, jambonz turns them on if alignment is enabled.

user_instructions

string

Spoken-delivery instructions for mars-instruct. Not used when streaming.

stability

number

Voice stability, 0 – 1. Not used when streaming.

temperature

number

Sampling temperature, 0.01 – 4. Not used when streaming.

speaker_similarity

number

How closely the output follows the voice, 0 – 1. Not used when streaming.

localize_speaker_weight

number

Weight of the localized speaker when the voice speaks another language, 0 – 1. Not used when streaming.

acoustic_quality_boost

boolean

Boost acoustic quality. Not used when streaming.

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

Camb.ai rejects text shorter than three characters, so a very short say such as "Hi" fails.

Endpoint override

To point at a non-default endpoint, such as a proxy, set the API base URL on the credential, or api_uri in its Extra Options. Leave it out to use client.camb.ai.

Streaming

Camb.ai 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 Camb.ai over a live connection with the mars-8.1-flash-beta model and opens a fresh connection for each turn; on barge-in it closes the connection and reconnects.

Camb.ai returns word-level timestamps, so alignment events are available when alignment is enabled.