Credits and licenses
A Fire-Press toolVox-Slice stands on other people's open work. Here is everything it uses and the license each piece comes with. Speech recognition runs on your computer: the model is downloaded once from Hugging Face and your audio is never uploaded.
Speech models
| Name | Used for | License |
|---|---|---|
| Xenova/whisper-base.en | Speech recognition, the default model (about 75 MB) | Apache-2.0 (Xenova ONNX conversion of OpenAI Whisper, MIT) |
| Xenova/whisper-tiny.en | Speech recognition, the fast model (about 40 MB) | Apache-2.0 (Xenova ONNX conversion of OpenAI Whisper, MIT) |
| Xenova/whisper-small.en | Speech recognition, the accurate model (about 250 MB) | Apache-2.0 (Xenova ONNX conversion of OpenAI Whisper, MIT) |
| OpenAI Whisper | The original speech recognition model the three above are converted from | MIT |
Libraries
| Name | Used for | License |
|---|---|---|
| Transformers.js (@huggingface/transformers) | Runs the Whisper models in your browser | Apache-2.0 |
| ONNX Runtime Web | The engine that executes the models (WebAssembly) | MIT |
| @breezystack/lamejs | Encodes MP3 files | LGPL-3.0 (copyleft, see below) |
| JSZip | Builds the ZIP download | MIT (offered as MIT or GPL-3.0-or-later; Vox-Slice uses it under MIT) |
| cmu-pronouncing-dictionary | Pronunciations for the rhyme finder | ISC (data derived from the CMU Pronouncing Dictionary, BSD-style) |
| Hugging Face jinja and tokenizers | Text handling inside Transformers.js | MIT and Apache-2.0 |
| protobufjs, flatbuffers, long, platform and small helpers | Pulled in by ONNX Runtime Web and JSZip (pako, readable-stream and similar) | MIT, BSD-3-Clause, ISC, Apache-2.0, Zlib |
| Vite | Build tool (not shipped to your browser) | MIT |
| Playwright | Browser testing (not shipped to your browser) | Apache-2.0 |
Fonts
| Name | Used for | License |
|---|---|---|
| Bebas Neue | Titles and the VOX-SLICE look | SIL OFL 1.1 |
| Inter | Body text and controls | SIL OFL 1.1 |
| Orbitron | Numbers and timecodes | SIL OFL 1.1 |
Artwork and code
| Name | Used for | License |
|---|---|---|
| Vox-Slice logo, monogram, app icons, background and social card | Brand artwork from the Fire-Press Vox-Slice asset pack v1.0 | © Fire-Press, all rights reserved. The code is MIT; the artwork is not. |
| Vox-Slice source code | This app | MIT |
Copyleft, in plain English
One copyleft piece is used: @breezystack/lamejs (LGPL-3.0), the MP3 encoder. The LGPL is a "weak" copyleft license. You can use Vox-Slice and the MP3s it makes freely, including commercially: the license covers the encoder's code, not your audio. The condition is that if anyone changes the encoder itself and shares that version, they must share their changes under the same license, and anyone who ships the encoder must let people swap in their own copy of it. Vox-Slice is fully open source (MIT) and pulls the encoder in as an ordinary, unmodified package, so anyone can rebuild the app with a different or updated encoder.
JSZip is offered under "MIT or GPL-3.0-or-later" and Vox-Slice takes the MIT option, so the strong GPL copyleft does not apply. No other GPL or copyleft dependency is used.
This is a good-faith summary, not legal advice. The license texts themselves are what count.
