# Live mode

<!-- WARNING: THIS FILE WAS AUTOGENERATED! DO NOT EDIT! -->

Live mode provides real-time transcription for voice agents, live captions, and other interactive audio applications. Your app streams audio over a WebSocket connection and receives JSON transcription events as speech is detected and processed.

## Capabilities

Live mode includes:

-   Continuous audio streaming over a secure WebSocket connection
-   A completed transcript segment after each detected speech turn
-   JSON events for session, audio-buffer, and transcription activity

## Connect to Live mode

To connect to Live mode, open a WebSocket connection to this endpoint.

**`wss://api.zoom.us/v2/aiservices/scribe/live`**

Then, include the `live-asr` WebSocket subprotocol and a [Zoom AI Services JWT](/docs/ai-services/auth/) in the `Authorization` header.

```javascript
import WebSocket from "ws";

const socket = new WebSocket(
    "wss://api.zoom.us/v2/aiservices/scribe/live",
    ["live-asr"],
    {
        headers: {
            Authorization: `Bearer ${process.env.ZOOM_ACCESS_TOKEN}`,
        },
    },
);
```

> **Note**
>
> Create the WebSocket connection from a trusted backend. Browser WebSocket clients cannot set the `Authorization` header, and placing a JWT in client-side code exposes your credentials. For browser audio capture, send audio through a backend WebSocket proxy that adds the authorization header.

## Configure the session

After the WebSocket opens, configure the session by sending a `session.update` event.

```json
{
    "type": "session.update",
    "language": "en-US",
    "audio": {
        "format": "pcm16"
    }
}
```

### Session options

| Parameter      | Type   | Description                                                                                                      |
| -------------- | ------ | ---------------------------------------------------------------------------------------------------------------- |
| `type`         | string | Message type. Set to `session.update`.                                                                           |
| `language`     | string | Transcription locale, such as `en-US`. See [supported languages](/docs/ai-services/scribe/#supported-languages). |
| `audio.format` | string | Audio encoding. Set to `pcm16`.                                                                                  |

## Audio

Clients stream binary PCM16 frames after configuring the session.

Use little-endian, 16 kHz, mono audio and stream it in approximately 100 ms frames.

Keep control messages, such as `session.update` and `session.close`, as JSON text messages. Do not wrap binary audio in JSON or Base64-encode it.

For browser microphone capture, an `AudioWorklet` can convert the browser's `Float32` samples to PCM16 frames before sending them to your backend proxy.

## Handle server events

The server returns JSON events throughout the session, including:

| Event                               | Payload fields besides `type`                                                         |
| ----------------------------------- | ------------------------------------------------------------------------------------- |
| `session.created`                   | `session_id`                                                                          |
| `session.updated`                   | None                                                                                  |
| `input_audio_buffer.speech_started` | `item_id`, `audio_start_ms`                                                           |
| `input_audio_buffer.speech_stopped` | `item_id`, `audio_end_ms`                                                             |
| `transcription.completed`           | `item_id`, `transcript`, `audio_start_ms`, `audio_end_ms`, `transcription_latency_ms` |
| `error`                             | `error: { code, message, fatal }`                                                     |
| `session.closed`                    | `reason`                                                                              |

This example sends the session configuration and handles transcription events:

```javascript
socket.on("open", () => {
    socket.send(
        JSON.stringify({
            type: "session.update",
            language: "en-US",
            audio: {
                format: "pcm16",
            },
        }),
    );
});

socket.on("message", (data) => {
    const event = JSON.parse(data.toString());

    if (event.type === "transcription.completed") {
        console.log("Final:", event.transcript);
    }

    if (event.type === "error") {
        console.error(event.error);
    }
});
```

## Close the session

To finish transcription and close the session.

1. Stop sending audio.
2. Send a `session.close` event:

```json
{
    "type": "session.close"
}
```

3. Continue reading events until the server sends any remaining final transcription and the `session.closed` event.
4. Close the WebSocket.

## Rate limits

Live mode rate limits apply per account.

### Concurrent sessions

| Account type         | Concurrent sessions |
| -------------------- | ------------------- |
| Default              | 20                  |
| Enterprise           | 100                 |
| Large Contact Center | 1,000               |
| Custom               | 5,000+              |

### Session limits

| Limit                    | Default                  |
| ------------------------ | ------------------------ |
| Maximum session duration | 60 minutes               |
| Idle timeout             | 30 seconds without audio |