Earkeep
Languages and models
Earkeep transcribes on your device with Whisper (99+ languages) or Parakeet-TDT v3 (25 European languages, including Czech, and the faster of the two), hardware-accelerated on Apple Silicon. Check your language in the list below before you buy.
Which model should I use
Mostly English: Parakeet-TDT v3. It tops the open ASR leaderboard on speed and beats Whisper large-v3 on English accuracy.
Another language, or a mix: if Parakeet covers it, Parakeet is faster there too. Otherwise use Whisper. It is slower for a given size but still transcribes as people speak; what you give up is headroom, not the live transcript.
The model is a setting you can change any time in Settings → Transcription or the left rail. The languages it listens for sit on the model's card: one, several, or auto-detect.
Model comparison
| Model | Size | What it's for | Approximate download |
|---|---|---|---|
| Whisper | tiny | Fastest and smallest, least accurate | 75 MB |
| Whisper | base | A step up from tiny | 142 MB |
| Whisper | small | Where accuracy starts being good enough to trust | 466 MB |
| Whisper | medium | More accurate, noticeably slower | 1.5 GB |
| Whisper | large-v3-turbo (default) | Close to large-v3 on accuracy, several times faster | 1.6 GB |
| Whisper | large-v3 | The family's accuracy ceiling, especially for mixed-language speech | 3.1 GB |
| Parakeet-TDT v3 | One size (int8-quantized) | 25 European languages, fastest on the open ASR leaderboard, beats large-v3 on English | 670 MB |
Your own Whisper model
People publish Whisper models fine-tuned for one dialect or field, such as Swiss German, which stock Whisper often hears as Dutch. Earkeep can run those too.
- Paste a Hugging Face link in Settings → Transcription → Add a model…. Earkeep shows which versions exist and how big each is before downloading anything.
- Transformers checkpoints are converted on your device to whisper.cpp format. PyTorch
.binfiles and LoRA adapters can't be converted. - Or choose a whisper.cpp
.binfile you already have. - The licence is shown before you download; whether it fits your use is between you and its author.
An added model works like the built-in ones. Only Whisper models can be added, not Parakeet.
On-device, offline, downloaded once
Inference runs on your device, so audio never travels anywhere to be transcribed and transcription works with no network. Each model downloads once and runs offline after that. See private by design for the full mechanism.
The complete language list
The 99 languages Whisper transcribes, at every size. The 25 in bold are also covered by Parakeet-TDT v3, the faster of the two.
- Afrikaans
- Albanian
- Amharic
- Arabic
- Armenian
- Assamese
- Azerbaijani
- Bashkir
- Basque
- Belarusian
- Bengali
- Bosnian
- Breton
- Bulgarian
- Burmese
- Catalan
- Chinese
- Croatian
- Czech
- Danish
- Dutch
- English
- Estonian
- Faroese
- Finnish
- French
- Galician
- Georgian
- German
- Greek
- Gujarati
- Haitian Creole
- Hausa
- Hawaiian
- Hebrew
- Hindi
- Hungarian
- Icelandic
- Indonesian
- Italian
- Japanese
- Javanese
- Kannada
- Kazakh
- Khmer
- Korean
- Lao
- Latin
- Latvian
- Lingala
- Lithuanian
- Luxembourgish
- Macedonian
- Malagasy
- Malay
- Malayalam
- Maltese
- Maori
- Marathi
- Mongolian
- Nepali
- Norwegian
- Nynorsk
- Occitan
- Pashto
- Persian
- Polish
- Portuguese
- Punjabi
- Romanian
- Russian
- Sanskrit
- Serbian
- Shona
- Sindhi
- Sinhala
- Slovak
- Slovenian
- Somali
- Spanish
- Sundanese
- Swahili
- Swedish
- Tagalog
- Tajik
- Tamil
- Tatar
- Telugu
- Thai
- Tibetan
- Turkish
- Turkmen
- Ukrainian
- Urdu
- Uzbek
- Vietnamese
- Welsh
- Yiddish
- Yoruba
Related
- Private by design: why audio never leaves your computer, and never touches disk, for either model.
- Always-on capture: why a fast model matters when transcription runs all day.
Frequently asked questions
Does Earkeep support my language?
Almost certainly, through Whisper's 99+ languages (list above). If it is one of Parakeet's 25, you can choose Parakeet for extra speed.
Can I switch models after installing?
Yes, any time in Settings → Transcription. The new model downloads the first time you switch to it.
Can I use a Whisper model fine-tuned for my dialect?
Yes, in most cases: paste its Hugging Face link in Settings → Transcription → Add a model…, or choose a whisper.cpp .bin file. See your own Whisper model above.
What accuracy should I expect?
It depends on the model. The default, large-v3-turbo, stays close to large-v3, Whisper's accuracy ceiling, while decoding several times faster. Parakeet-TDT v3 beats large-v3 on English but covers only 25 languages.
How big are the model downloads?
Whisper ranges from 75 MB (tiny) to 3.1 GB (large-v3); Parakeet-TDT v3 is about 670 MB. Each downloads once and then runs offline.
Does my audio go anywhere during transcription?
No. Audio is processed only in memory on your device, never uploaded and never written to disk.
Find your language, then try it
14-day full trial, no account required.