BektavבכתבBektav — Third-party notices
Bektav and tScribe are two names of the same Service, provided by the same company from the same systems: an account made under either name is one account, kept in one database and managed by the same administrators; this document applies to both.
Last updated: 29 October 2026
Bektav is built with open-source software and openly licensed models. We thank their authors. Each is used under its own licence, as listed below.
Models
| Component | Use in Bektav | Licence |
|---|---|---|
| WeSpeaker ResNet34, trained on VoxCeleb (WeSpeaker project; VoxCeleb dataset by the University of Oxford VGG) | telling voices apart, voice prints | CC BY 4.0 |
| Whisper large-v3 and large-v3-turbo (OpenAI), CTranslate2 conversions | speech recognition | MIT |
| ivrit.ai whisper-large-v3 and whisper-large-v3-turbo (CTranslate2) | Hebrew speech recognition | Apache 2.0 |
| pyannote segmentation-3.0 | finding who speaks when | MIT |
Tesseract tessdata_best | reading text in pictures | Apache 2.0 |
| MADLAD-400 3B MT (Google; CTranslate2 int8 conversion by Nextcloud) | the local translation | Apache 2.0 |
Software
| Component | Licence |
|---|---|
| Tesseract OCR | Apache 2.0 |
| faster-whisper, CTranslate2 | MIT |
| sherpa-onnx, ONNX Runtime | Apache 2.0 / MIT |
| FFmpeg (a GPL build, run as a separate program) | GPL |
| libheif, libde265 (in pillow-heif) | LGPL 3.0 |
| x265 (in pillow-heif) | GPL 2.0 |
| cysignals | LGPL 3.0 or later |
| certifi, tqdm | MPL 2.0 |
| FastAPI, Starlette, Uvicorn, NumPy, Pillow, PyMySQL, cryptography, pymorphy3, snowballstemmer and other Python libraries | MIT / BSD / Apache 2.0 |
| React, @tanstack/react-virtual, qrcode-generator | MIT |
Fonts and pictures
| Component | Licence |
|---|---|
| Inter, IBM Plex Sans Hebrew | SIL Open Font License 1.1 |
| Photos on this site, from Unsplash | Unsplash License |
The full licence texts come with each component. To ask for the source code of a GPL or LGPL component we use, write to [email protected].