> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.jambonz.org/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.jambonz.org/_mcp/server.

# Cartesia

Cartesia's tuning is passed in `synthesizer.options`. Which fields take effect depends on the model family, because Cartesia moved these controls between API versions.

```json
{
  "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"
    }
  }
}
```

## Options

**`speed`** `number`

Speaking rate. Defaults to `0.0`. Negative values are slower, positive faster.

---

**`emotion`** `string`

An emotion tag in `name:level` form — for example `positivity:high`. Supported names are `anger`, `positivity`, `surprise`, `sadness` and `curiosity`; levels are `lowest`, `low`, `high` and `highest` (not every combination is offered for every emotion).

---

**`volume`** `number`

Output volume. Applies to the `sonic-3` family only.

---

**`max_buffer_delay_ms`** `number`

How long Cartesia may buffer incoming text before synthesising, in milliseconds. Raising it can improve prosody across token boundaries at the cost of latency. Not exposed in the portal — set it on the verb.

---

## Model family differences

> **Warning**
>
> `speed`, `emotion` and `volume` are sent differently depending on the model, and a control sent to the wrong family is ignored:
>
> * **`sonic-3` and later** — sent as `generation_config`. All three of `speed`, `emotion` and `volume` apply.
> * **`sonic-2`** — sent as experimental voice controls. Only `speed` and `emotion` apply; `volume` has no effect.

If a speed or emotion setting appears to do nothing, check which model the credential is using before assuming the value is wrong.

## Streaming

Cartesia supports [TTS streaming](/guides/features/tts-streaming), including a native cancel for barge-in and word-level timestamps for alignment.