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

# Transcription models

> Steno transcribes on-device with one of two engines: Parakeet (the default) and Whisper (optional). Choose the engine that matches your language and workflow needs.

Steno transcribes recordings entirely on your Mac -- no audio is sent to any server. There are two transcription engines: **Parakeet**, the default, and **Whisper**, an optional alternate you can switch to in Settings.

## The two engines

| Engine | Model | Size | Notes |
| - | - | - | - |
| **Parakeet** (default) | `mlx-community/parakeet-tdt-0.6b-v3` | \~572MB | Live transcript while recording; 25 European languages, including English |
| **Whisper** (optional) | `large-v3-turbo` | \~1.6GB | Full 99-language coverage; transcribes after you stop |

**Parakeet** is the default on new installs. It runs via MLX on Apple Silicon and shows a live transcript as you record.

**Whisper** is a single model, `large-v3-turbo`. Choose it in **Settings → Transcribe** when you need a language Parakeet does not cover. Whisper recordings are transcribed only after you stop the recording.

Both engines run 100% on-device.

## How to change engines

1. Open **Settings** in Steno
2. Select **Transcribe**
3. Choose **Parakeet** or **Whisper**
4. If the engine's model is not yet downloaded, Steno will prompt you to download it

The engine is applied to all future recordings. Existing recordings are not re-transcribed automatically -- transcription only runs once, when the recording is first processed.

## Speed

Transcription runs on-device and is fast on Apple Silicon. Parakeet additionally shows a live transcript while you record, so notes are ready shortly after you stop. Whisper transcribes after the recording ends rather than during it.

## Language support

Language coverage depends on the engine you choose.

* **Parakeet (default)** transcribes 25 European languages, including English, French, German, Spanish, Italian, Portuguese, Dutch, Russian, and Polish. A curated subset of six (English, French, German, Spanish, Dutch, Portuguese) can be pinned as the summary output language.
* **Whisper (optional)** covers the full set of **99 languages**, auto-detected. Switch to Whisper for Japanese, Chinese, Korean, Hindi, Arabic, and any language outside Parakeet's set.

Steno auto-detects the language from the audio -- you do not need to configure this manually.

## Accuracy tips

* **Record in a quiet environment** -- background noise is the biggest accuracy factor
* **Use system audio capture** -- capturing both call sides is more accurate than microphone-only for virtual meetings
* **Switch to Whisper for non-European languages** -- Parakeet covers 25 European languages, including English; Whisper covers the full 99-language set

***

<Accordion title="Which engine should I start with?">
  Start with **Parakeet** (the default). It runs on-device, shows a live transcript while you record, and covers 25 European languages, including English. Switch to **Whisper** in **Settings → Transcribe** if you need a language outside Parakeet's set.
</Accordion>

<Accordion title="Can I switch engines for future recordings?">
  Yes. Change the engine in **Settings → Transcribe** -- it applies to recordings from that point on. There is currently no way to re-transcribe an already-processed recording with a different engine; you can regenerate its summary, but not its transcript.
</Accordion>

<Accordion title="Does transcription accuracy vary by language?">
  Yes. Parakeet covers 25 European languages, including English, French, German, Spanish, Italian, Dutch, Portuguese, Russian, and Polish. For other languages -- including Japanese, Chinese, Korean, Hindi, and Arabic -- switch to the Whisper engine, which covers all 99 languages.
</Accordion>


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