Transcription Settings

Transcription settings control how Yapper converts your speech to text. Here you can choose your transcription engine and model, set your language, manage model storage, and add custom vocabulary for better recognition.

Accessing Transcription Settings

Click the Yapper icon in the menu bar
Select Settings (or press ⌘ ,)
Go to the Transcription tab

Transcription Engines

Yapper supports two speech recognition engines, both running entirely on your Mac:

Parakeet (NVIDIA) — Very fast, excellent accuracy, supports 25 EU languages. Recommended for most users.
Whisper (OpenAI) — Broad language support (~100 languages), supports streaming transcription.

Both engines process speech locally — your voice never leaves your computer.

Speech Models

Models are grouped by engine in the Settings picker. Select a model to use it for all transcription.

Parakeet Models (Recommended)

Model	Accuracy	Speed	Storage	Languages
Parakeet TDT v3 (Default)	Excellent	Very Fast	~800 MB	25 EU languages
Parakeet TDT v2	Excellent	Very Fast	~800 MB	English only

Parakeet models use NVIDIA’s TDT architecture and run at 110-190x real-time speed — meaning a 10-second recording is transcribed almost instantly.

Whisper Models

Model	Accuracy	Speed	Storage	Languages
Large v3 Turbo	Excellent	Fast	~1.5 GB	~100 languages
Large v3	Excellent	Slower	~3 GB	~100 languages
Small	Good	Faster	~500 MB	~100 languages
Base	Moderate	Very Fast	~250 MB	~100 languages
Tiny	Basic	Fastest	~150 MB	~100 languages

Whisper models run on Apple’s Neural Engine. They support the widest range of languages including Arabic, Chinese, Hindi, Japanese, Korean, Thai, Vietnamese, and many more.

Choosing a Model

Parakeet TDT v3 is the default and recommended choice — it offers excellent accuracy with the fastest processing. Consider other models if:

You need a non-EU language (e.g., Chinese, Japanese, Arabic) → Use a Whisper model (Large v3 Turbo recommended)
You want streaming transcription → Use a Whisper model (text appears word-by-word as you speak)
You need maximum accuracy across all languages → Large v3
You want English-only with top accuracy → Parakeet TDT v2
You have limited storage → Small (~500 MB, still good accuracy)

Downloading Models

Models are downloaded on first use:

Select a model from the dropdown
If it’s not already downloaded, a progress bar appears
Wait for the download to complete — or cancel anytime by clicking the cancel button next to the progress bar
The model is now ready to use

Downloads require an internet connection. Once downloaded, models work entirely offline.

Model Loading

When you open Yapper or change models, the selected model needs to load into memory. You’ll see:

Loading indicator — Model is being loaded
Ready — Model is loaded and ready for transcription

If you try to record before the model is ready, Yapper will wait until loading completes.

Model Storage

The Transcription settings tab shows how much disk space your downloaded models are using.

Clearing Downloaded Models

To free disk space, click Clear Downloaded Models in the Transcription settings. This removes all downloaded models from both engines. The currently selected model will be re-downloaded automatically when needed.

Models are stored locally at:

Whisper models: ~/Library/Application Support/Yapper/models/
Parakeet models: ~/Library/Application Support/Yapper/parakeet-models/

Language

Select the language you’ll be speaking. Available languages depend on the selected model:

Parakeet TDT v3 supports 25 EU languages: Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish, Ukrainian

Parakeet TDT v2 supports: English only

Whisper models support ~100 languages including: Arabic, Chinese, English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Polish, Portuguese, Russian, Spanish, Thai, Turkish, Ukrainian, Vietnamese, and many more

The language picker automatically filters to show only languages supported by the currently selected model. If you switch to a model that doesn’t support your previously selected language, it will reset to English.

Setting Your Language

Click the Language dropdown
Select your language
The setting saves automatically

Language Tips

Match your spoken language — Always set the language you’ll actually be speaking
Regional variations — Choose the base language (e.g., “English” works for US, UK, Australian English)
Switching engines? — If your language isn’t supported by your new model, Yapper resets to English automatically

Dual-Language Switching

If you regularly speak in two languages, you can set up a secondary language and switch between them instantly — no need to open Settings each time.

Setting Up a Secondary Language

Go to the Transcription tab in Settings
Set your Primary Language (the one you use most)
Set a Secondary Language from the dropdown

Switching Languages

Once a secondary language is set, toggle between your two languages with the global hotkey:

⇧ ⌥ L (Shift + Option + L)

A floating pill appears briefly on screen, showing the flag of the newly active language to confirm the switch.

How It Works

Primary language is active by default when Yapper launches
Press the language toggle hotkey to switch to your secondary language
Press it again to switch back to your primary language
The active language persists until you switch again or restart Yapper

Customizing the Hotkey

You can change the language toggle shortcut in Settings → Preferences → Keyboard Shortcuts.

Custom Vocabulary

Custom vocabulary helps Yapper recognize words it might otherwise miss—proper nouns, technical terms, acronyms, and uncommon words.

Why Use Custom Vocabulary

The speech model doesn’t know about:

Your name or colleagues’ names
Your company name
Industry-specific terms
Acronyms you use frequently

Adding these to custom vocabulary improves recognition accuracy.

Adding Words

Find the Custom Vocabulary section
Type a word or phrase in the “Add word or phrase…” field
Press Enter or click Add
The word appears as a tag below the input

Examples of Good Custom Vocabulary

Category	Examples
Names	Your name, colleagues, clients
Companies	Your company name, competitors, partners
Products	Your product names, features
Technical terms	API, OAuth, Kubernetes, YAML
Acronyms	CRM, SQL, AWS, CEO, EOD
Industry terms	Specific jargon for your field

Removing Words

Click the X on any vocabulary tag to remove it.

Best Practices

Add words you actually use — Don’t add every technical term, just ones you say frequently
Include proper capitalization — Add “McKenzie” not “mckenzie”
Add common mistakes — If a word is consistently misheard, add the correct version
Keep it manageable — A focused list of 20-50 words is more effective than hundreds

Settings Persistence

All transcription settings are saved automatically and persist across app restarts. Settings are stored locally on your Mac in your preferences.

Troubleshooting

Model won’t download

Check your internet connection
Ensure you have enough free disk space
Cancel and retry the download
Try a smaller model first
Restart Yapper and try again

Model takes long to load

Larger models take more time to load
First load after download is slower
Subsequent loads are faster due to caching

Poor transcription accuracy

Try a different model — Parakeet TDT v3 is recommended for EU languages, Large v3 Turbo for other languages
Verify the language setting matches what you’re speaking
Add frequently misheard words to custom vocabulary
Check your microphone input quality

For more help, see Model Issues.

Live Speech-to-Text — Using transcription
Model Issues — Troubleshoot model problems
General Settings — Other settings