Minutify audio cleanup engine: third-party files
================================================

This folder holds the files the audio cleanup page downloads once a visitor starts a run,
plus the sources and licences for them. Nothing here is loaded before that.

Each file below has its version, where it comes from, its licence and its SHA-256.


1. FFmpeg (ffmpeg.wasm core, LGPL build)
----------------------------------------

Files:
SHA-256 962d19bf543408b66eeb500c5cfef011c022d1ba69e1c6263ef2d327d4b428bb  ffmpeg-core.mjs
SHA-256 c46317227cb5189c3df55c9b46b125ba78935d82ca4db61b981be7c2cf903e92  ffmpeg-core.wasm

Version:     FFmpeg 5.1.10
Source:      https://ffmpeg.org/releases/ffmpeg-5.1.10.tar.xz
Source SHA-256: 392306d6fc45dab0e9e0ea55381e071842e83a2fb31d320aeda40477a7766293
Compiler:    Emscripten 6.0.13 (Docker image emscripten/emsdk:6.0.13)
Licence:     GNU Lesser General Public License 2.1 or later (LGPL). Built without
             --enable-gpl and without --enable-nonfree, so no GPL or non-free code is in it.
             Licence text: LICENSE-ffmpeg-LGPL-2.1.txt

The core is single-thread. FFmpeg 6.1 and later need threads for the ffmpeg program, so 5.1,
a release line FFmpeg still patches, is the newest that builds without them. It reads m4a, mp3
and wav (AAC, ALAC, MP3 and PCM audio) and writes AAC audio in an m4a file.

Build recipe and flags:
SHA-256 522f550e5ef0e1363ab95c1f7f88c0c29c8216166f62cf85fa2b66fb6d749a13  build_ffmpeg_wasm.sh.txt
SHA-256 23c414c54b4f7e0aecfec7cccb2761fdb140c57b87c74f04438c967c1b8577d6  ffmpeg-configure.txt
SHA-256 246041b6ecf9bc32d718a62c57877c78b5eb397b6467e74ed7ae2626ab189c30  LICENSE-ffmpeg-LGPL-2.1.txt
SHA-256 620a78084fc7ca97c0b5dea9abf891f3ffcadfdbf305276f099c9c4e12fc1d86  LICENSE-emscripten.txt

The Emscripten runtime and the musl C library compiled into the core are MIT (Emscripten's
own licence also offers NCSA): LICENSE-emscripten.txt, from
https://raw.githubusercontent.com/emscripten-core/emscripten/6.0.13/LICENSE

To rebuild the core, with Docker installed:

    mkdir -p rebuild/tools rebuild/website/src/tools/vendor
    cp build_ffmpeg_wasm.sh.txt rebuild/tools/build_ffmpeg_wasm.sh
    bash rebuild/tools/build_ffmpeg_wasm.sh

The script downloads the FFmpeg source above, checks its SHA-256, runs FFmpeg's configure with
the flags below inside the Emscripten image, and writes into rebuild/website/src/tools/vendor/:
ffmpeg-core.mjs, ffmpeg-core.wasm, ffmpeg-configure.txt, the script copy and the LGPL text.
The build is reproducible: the same inputs give the same bytes.

Configure flags (same as ffmpeg-configure.txt):

    --target-os=none
    --arch=x86_32
    --enable-cross-compile
    --disable-asm
    --disable-x86asm
    --disable-inline-asm
    --disable-pthreads
    --disable-w32threads
    --disable-os2threads
    --disable-autodetect
    --disable-runtime-cpudetect
    --disable-network
    --disable-debug
    --disable-doc
    --disable-stripping
    --disable-everything
    --disable-programs
    --enable-ffmpeg
    --disable-avdevice
    --disable-swscale
    --disable-postproc
    --enable-protocol=file
    --enable-demuxer=mov,mp3,wav
    --enable-parser=aac,mpegaudio
    --enable-decoder=aac,alac,mp3float,pcm_s16le,pcm_s16be,pcm_s24le,pcm_s24be,pcm_s32le,pcm_s32be,pcm_f32le,pcm_f32be,pcm_f64le,pcm_u8,pcm_alaw,pcm_mulaw
    --enable-encoder=aac,pcm_s16le
    --enable-muxer=ipod,pcm_s16le
    --enable-filter=abuffer,abuffersink,anull,aformat,aresample,highpass,acompressor,loudnorm,asetnsamples,aselect,asetpts,atempo,asplit
    --cc=emcc
    --cxx=em++
    --ar=emar
    --ranlib=emranlib
    --nm=emnm


2. ONNX Runtime Web (runs the voice detector)
---------------------------------------------

Files:
SHA-256 51bd7168175767324ab0acfcfc4ca94a096a3826e3ebac9ccc4f195eb865f8dd  ort.wasm.bundle.min.mjs
SHA-256 4b884c203b0fa48d057be3332c65513ff85aa3c5bddd6946acc0fab7ecda001d  ort-wasm-simd-threaded.wasm
SHA-256 2f07c72751aed99790b8a4869cf2311df85a860b22ded05fa22803587a48922c  LICENSE-onnxruntime.txt
SHA-256 143764b952fdb1a7c69ce653bfba74a7744d6a8a573bfb73e235fba356c83de3  ThirdPartyNotices-onnxruntime.txt

Version:     onnxruntime-web 1.31.1
Source:      https://registry.npmjs.org/onnxruntime-web/-/onnxruntime-web-1.31.1.tgz
             (files dist/ort.wasm.bundle.min.mjs and dist/ort-wasm-simd-threaded.wasm,
             unchanged), project https://github.com/microsoft/onnxruntime
Licence:     MIT. Licence text: LICENSE-onnxruntime.txt. The wasm build links third-party code
             (protobuf, abseil, ONNX, Eigen and others); their notices are in
             ThirdPartyNotices-onnxruntime.txt, from
             https://raw.githubusercontent.com/microsoft/onnxruntime/v1.31.1/ThirdPartyNotices.txt
Runs with one thread (numThreads = 1).


3. Silero VAD (voice activity detector model)
---------------------------------------------

Files:
SHA-256 7ed98ddbad84ccac4cd0aeb3099049280713df825c610a8ed34543318f1b2c49  silero_vad_16k_op15.onnx
SHA-256 2e63e9a38b6e8fc0c7bc37ce174caca1862870856c6daf5697cfb785e925520b  LICENSE-silero-vad.txt

Version:     Silero VAD v6.2.3
Source:      https://raw.githubusercontent.com/snakers4/silero-vad/v6.2.3/src/silero_vad/data/silero_vad_16k_op15.onnx
             (unchanged), project https://github.com/snakers4/silero-vad
Licence:     MIT. Licence text: LICENSE-silero-vad.txt