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

# Speech to text overview

> Transcribe English and Indian-language speech with Chariot, live over a streaming WebSocket.

Chariot speech to text (ASR) turns spoken audio into text. It is built for English and Indian languages, including speech that mixes the two in one sentence.

You transcribe over a streaming WebSocket:

| Transport | Endpoint | Input | Output | Best for |
| - | - | - | - | - |
| **Streaming WebSocket** | `wss://api.chariot.in/v1/asr/ws` | Raw audio frames, sent as you capture them | Text as it is recognized, then a final transcript | Voice agents, live captions, phone calls, dictation, recordings |

It accepts microphone, telephony, and recorded audio, and handles sessions of up to three hours. See the [quickstart](/guides/speech-to-text/quickstart).

## Model

| Model | `model_type` | Transports |
| - | - | - |
| Chariot 1 ASR Realtime | `chariot-1-asr-realtime` | Streaming WebSocket (default and only value) |

You can omit `model_type`. See [Models](/guides/models).

## What you can control

Streaming sessions take these options on the connection URL. Each one is fixed for the session.

| Option | Values | Default | What it does |
| - | - | - | - |
| `language` | `en`, `hi`, `bn`, `gu`, `kn`, `ml`, `mr`, `or`, `pa`, `sa`, `ta`, `te`, or `none` | Automatic language detection | Tells the model which language to expect, or `none` for automatic language detection. It does not translate. |
| `mode` | `balanced`, `high_accuracy` | `balanced` | Trades latency for accuracy. |
| `style` | `codemix`, `native` | `codemix` | Chooses the script for borrowed words in mixed-language speech. |
| `audio_format` | `pcm_8000` to `pcm_48000`, `ulaw_8000`, `alaw_8000` | `pcm_16000` | Declares the encoding and sample rate of the audio you send. |

See [Languages, modes, and styles](/guides/speech-to-text/modes-and-languages) and [Audio formats](/guides/speech-to-text/audio-formats).

## What you get back

A streaming session sends `transcript.delta` events with new text as it is recognized, then one `transcript.done` event with the complete transcript.

Transcripts are plain text. Word-level timestamps, confidence scores, and speaker labels are not part of the response.

## Cost

Speech to text costs **1.5 credits per second of audio** (90 credits per minute). Streaming bills the audio you send, not the time the connection is open. See [Credits & pricing](/guides/resources/credits).

## History

Every streaming session is saved to your workspace with its transcript, settings, and cost. List them with [`GET /v1/asr/sessions`](/api-reference/endpoint/asr-sessions). See [Session history](/guides/speech-to-text/history).

## Next steps

> **[Quickstart](/guides/speech-to-text/quickstart)**
>
> Stream a file and get a transcript in a few minutes.

> **[Streaming WebSocket](/guides/speech-to-text/websocket)**
>
> The full real-time protocol.

> **[Audio formats](/guides/speech-to-text/audio-formats)**
>
> Sample rates, telephony audio, and chunk sizes.

> **[Session history](/guides/speech-to-text/history)**
>
> Look up past sessions and their transcripts.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.