Licenses

Every component below is served from this site's own domain. Nothing here is loaded from a CDN, and no component is fetched from a third party at runtime — the licences are listed on this page because the files themselves are distributed from here.

The full text of each licence is published next to the component, not reproduced in this page. Where a component was built for this site from someone else's source, the statement of provenance and the link to that source is in the same folder as the licence text.

The video-extraction page runs ffmpeg-core 0.12.10, a WebAssembly build of FFmpeg distributed under the GNU General Public License, version 2 or later. The full licence text is at /ffmpeg/LICENSE.GPL-2.0.txt. The binary is served as four byte-range parts and reassembled in the browser; the parts are the unmodified upstream build, not a recompiled one.

Source and provenance: /ffmpeg/SOURCE.txt names the exact upstream release and links to the corresponding source of both the build and FFmpeg itself. The JavaScript wrapper, @ffmpeg/ffmpeg 0.12.15, is a separate package under the MIT licence.

The two bitmap-subtitle pages read the text out of PGS and VobSub images with tesseract.js 7.0.0, which is distributed under the Apache License 2.0. The full text is at /tesseract/LICENSE.Apache-2.0.txt, and /tesseract/SOURCE.txt records the exact version and where the files came from.

Its recognition models — the fast traineddata files for the languages the two pages offer — are separate MIT-licensed data packages. Only the model for the language you pick is downloaded, and the request carries nothing from your file.

The transcription pages run Whisper, exported to ONNX by the onnx-community project from OpenAI's original weights. The weights are distributed under the MIT licence, and the full text is at /models/LICENSE.MIT.txt. Two sizes are published, tiny and base; only the one you pick is downloaded.

The narration pages run Kokoro-82M, distributed under the Apache License 2.0, full text at /models/LICENSE.Apache-2.0.txt. Both model sets are served from this domain as parts smaller than the hosting limit and reassembled in the browser; /models/SOURCE.txt records each upstream repository, its pinned commit and the file counts. No model is fetched from a third party at runtime.

Both kinds of model run on onnxruntime-web, distributed under the MIT licence. The WebAssembly core is served from /ort/, with its licence at /ort/LICENSE.MIT.txt and its version in /ort/SOURCE.txt. It is a separate file, not bundled into the page's JavaScript.

Turning text into the sounds a voice model can pronounce is done with eSpeak NG, a WebAssembly build distributed under the GNU General Public License, version 3. The full licence text is at /espeak/LICENSE.GPL-3.0.txt. It is served as separate files from /espeak/, not compiled into this site's own code, and the build is unmodified.

Source and provenance: /espeak/SOURCE.txt names the npm package, the pinned eSpeak NG commit and the upstream repository, and /espeak/THIRD_PARTY_NOTICES.md lists the components bundled inside that build. eSpeak NG is used to phonemise text only; which voice speaks the result is Kokoro's.

PDF output embeds a subset of Noto Sans JP, Noto Sans SC or Noto Sans KR, distributed under the SIL Open Font License 1.1. The licence files are at /fonts/OFL-jp.txt, /fonts/OFL-sc.txt and /fonts/OFL-kr.txt. A document that needs none of these scripts downloads none of them.

PDFs are written with pdf-lib and @pdf-lib/fontkit, both under the MIT licence. These are ordinary JavaScript modules bundled into a lazily loaded chunk; they are not separate files you download, and no font or binary is fetched from them.

Everything not listed above — the subtitle parsers, the capability matrix, the loss reports, the layout engine and the interface — is this site's own code. This page grants no licence to reuse it; using the tools through the browser is what it is for.