> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bland.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Migrate from ElevenLabs

> Move an ElevenLabs streaming TTS client to Bland's realtime WebSocket.

ElevenLabs and Bland both support a long-lived WebSocket for partial LLM text. The main wire change is that Bland sends raw binary audio and uses explicit turns instead of JSON responses with base64 audio.

Start with the working [Realtime TTS Quickstart](/tts/realtime-quickstart), then apply these mappings.

## Message mapping

| ElevenLabs stream-input                          | Bland `/v2/tts/ws`                                                |
| ------------------------------------------------ | ----------------------------------------------------------------- |
| Voice ID in the WebSocket URL                    | Bland voice UUID in the first `init` message                      |
| `xi-api-key` header                              | `Authorization: Bearer <BLAND_API_KEY>`                           |
| Initial `{ "text": " ", "voice_settings": ... }` | `{ "type": "init", "voice": ..., "audio": ..., "controls": ... }` |
| `{ "text": token }`                              | `{ "type": "speak", "context_id": turnId, "text": token }`        |
| Final empty text or flush behavior               | `{ "type": "end_of_turn", "context_id": turnId }`                 |
| JSON `audio` field containing base64             | Binary WebSocket frame containing raw audio                       |
| `is_final: true`                                 | `utterance_end` with `reason: "complete"`                         |

## Replace the send loop

```js theme={null}
const turnId = crypto.randomUUID();

for await (const token of llmTextStream) {
  ws.send(
    JSON.stringify({
      type: "speak",
      context_id: turnId,
      text: token,
    }),
  );
}

ws.send(JSON.stringify({ type: "end_of_turn", context_id: turnId }));
```

Do not port ElevenLabs chunk schedules, `try_trigger_generation`, or sentence timers. Bland buffers token input and chooses speech boundaries on the server.

## Controls and output

* Map ElevenLabs `stability` to Bland `controls.stability`, both from `0.0` to `1.0`.
* Tune Bland `controls.expressiveness` independently. `similarity_boost`, `style`, speaker boost, speed, and seed do not have direct Bland equivalents.
* Bland realtime output is `pcm_s16le` at 8, 16, 24, 44.1, or 48 kHz, or 8 kHz mu-law. It does not emit MP3 over this WebSocket.
* Bland does not currently return alignment or timestamp events.

On interruption, send the replacement response under a new `context_id`. Bland preempts the old turn automatically. Also discard any old audio that remains in your local playback queue.

See the [ElevenLabs WebSocket reference](https://elevenlabs.io/docs/api-reference/text-to-speech/v-1-text-to-speech-voice-id-stream-input) and the [Bland WebSocket reference](/api-v2/post/tts-ws).

***

Docs for agents: [llms.txt](/llms.txt)