Voice translator powered by neural networks — without the internet
Whisper transcribes speech, NLLB-200 translates, Piper speaks — all on your device, no cloud, no sign-up, no tracking. Download the models once and use them free forever.
What's inside
~1.9 GBDownload neural network models
Pick a source, download the
LetItTalk
folder and place it in
Documents/LetItTalk
on your device. The app will pick up the files automatically.
Developer siteRecommended
Direct download from the primary CDN. The built-in downloader uses this source automatically.
Used automatically by the app
Google Drive
Google Drive mirror. Use it if the main server is unavailable or a VPN blocks traffic.
Open folder
Yandex Disk
Yandex Disk mirror. Typically fast from Russia and CIS countries without a VPN.
Open folderFile list (8 files, ~1.9 GB)
| File | Size | Purpose |
|---|---|---|
whisper_encoder.onnx | ~87 MB | Whisper STT encoder |
whisper_decoder.onnx | ~155 MB | Whisper STT decoder |
whisper_tokenizer.json | ~2 MB | Whisper BPE tokenizer |
nllb_encoder.onnx | ~591 MB | NLLB-200 encoder |
nllb_decoder.onnx | ~917 MB | NLLB-200 decoder |
nllb_tokenizer.json | ~18 MB | NLLB SentencePiece tokenizer |
piper_model.onnx | ~61 MB | Piper TTS model (en-US, amy-medium) |
piper_config.json | ~5 KB | Piper config |
Support the project
The models are free and will stay free forever. If you find the app useful, you can buy a copy in the store — it helps the project grow.
Get the app in the store
Buy LetItTalk at ЧёКак.Store — support an indie developer.
About the developer
Ivan Kosenkov — engineer, author of this project. More work and contacts on the personal site.