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

# Gladia

> Use the Gladia speech-to-text plugin in a Zero Runtime pipeline. Setup and usage in Python.

Gladia is a **speech-to-text** plugin. It transcribes the caller's audio into text for
the LLM, with support for 90+ languages and automatic code-switching between them.

## Setup

Set your Gladia API key in the worker environment. Generate a key from the [Gladia dashboard](https://app.gladia.io/signup):

```bash theme={null}
export GLADIA_API_KEY=<key>
```

## Usage

Import the plugin and pass it to the pipeline's `stt` slot.

```python theme={null}
from zrt.plugins import GladiaSTT

stt = GladiaSTT()

# Pipeline(stt=stt, ...)
```

## Parameters

| Parameter                     | Type                | Default       | Description                                                                                                                                       |
| ----------------------------- | ------------------- | ------------- | ------------------------------------------------------------------------------------------------------------------------------------------------- |
| `api_key`                     | `str \| None`       | `None`        | Gladia API key. Falls back to the `GLADIA_API_KEY` environment variable when unset.                                                               |
| `model`                       | `str`               | `"solaria-1"` | Gladia recognition model. `"solaria-1"` is the only production model and the default.                                                             |
| `languages`                   | `List[str] \| None` | `None`        | ISO 639-1 language codes to hint at (e.g. `["en", "fr"]`). Only the first element is used. When `None`, the default is `"english"`.               |
| `code_switching`              | `bool`              | `True`        | When `True`, language is re-detected on every utterance. When `False`, language is detected once on the first utterance and held for the session. |
| `input_sample_rate`           | `int`               | `48000`       | Sample rate (Hz) of the incoming audio before any resampling. Valid values: `8000`, `16000`, `32000`, `44100`, `48000`.                           |
| `output_sample_rate`          | `int`               | `16000`       | Sample rate (Hz) forwarded to the Gladia WebSocket session.                                                                                       |
| `encoding`                    | `str`               | `"wav/pcm"`   | PCM encoding format sent over the WebSocket. Valid values: `"wav/pcm"`, `"wav/alaw"`, `"wav/ulaw"`.                                               |
| `bit_depth`                   | `int`               | `16`          | Bit depth of the PCM samples. Valid values: `8`, `16`, `24`, `32`.                                                                                |
| `channels`                    | `int`               | `1`           | Number of audio channels (1-8). Default is mono.                                                                                                  |
| `receive_partial_transcripts` | `bool`              | `False`       | When `True`, partial (non-final) transcripts are emitted over the WebSocket before the utterance is complete.                                     |

Transcribed text passes to the [LLM](/plugins/llm/google) once
[turn detection](/plugins/turn-detection/namo) decides the caller has finished speaking.