Third-party notices

FoxySing itself is proprietary software under the end-user license agreement in LICENSE. Everything below is not part of it: it is downloaded during installation (Python packages from PyPI, PyTorch from download.pytorch.org) or on first use (the separation models are optional: without them songs can be split only in the Cloud mode). Each component keeps its own license; using FoxySing means using them under those terms.

Summary — what this means for you

What the default models were trained on

A model's license covers its code and weights; the data it was trained on can have terms of its own (this is why the author of Demucs treats its weights as research-only). What the authors publish about the models FoxySing installs:

Some of these datasets are published for research or non-commercial use only. Whether that affects models trained on them is a legal question that has not been settled; it is listed here so that it can be checked.

RMVPE: where the file comes from

Programs downloaded by the installer (into .tools)

ProgramUsed forSourceLicense
uv 0.13.7installs FoxySing's own PythonAstral, github.com/astral-sh/uvMIT or Apache-2.0
CPython 3.14.10 (python-build-standalone)runs FoxySingPython Software Foundation; builds by AstralPSF License
FFmpeg, LGPL build N-128433 (Windows, only if FFmpeg is not on the computer)reading and decoding audio filesFFmpeg developers; Windows build by BtbN, github.com/BtbN/FFmpeg-Builds (pinned release autobuild-2026-11-05-12-48)LGPL-3.0 (license text in .tools\ffmpeg\LICENSE.txt); source code at ffmpeg.org
yt-dlp (PyPI wheel, the latest version on purpose: YouTube changes often and FoxySing updates yt-dlp itself when downloads break; the installed version is in the journal and in diagnose.txt) with yt-dlp-ejs (the version yt-dlp names), optionaldownloading audio from a YouTube link; kept in tools\yt-dlp; checksums checkedPyPI (yt-dlp, yt-dlp-ejs); source github.com/yt-dlpUnlicense (yt-dlp); Unlicense, MIT and ISC (yt-dlp-ejs)
Deno v2.9.8 (optional; pinned)runs the JavaScript that yt-dlp needs for YouTube; kept in tools\deno; checksum checkedgithub.com/denoland/denoMIT
cloudflared 2026.10.7 (optional)the «Sing together» link for the whole internet (Cloudflare Quick Tunnel); downloaded only after you agree, kept in tools\cloudflared; checksum checkedCloudflare, github.com/cloudflare/cloudflaredApache-2.0
Microsoft Visual C++ Redistributable 2015-2022 (x64) (only if missing or older than 14.38)runtime needed by Python and PyTorch on WindowsMicrosoft, downloaded from https://aka.ms/vs/17/release/vc_redist.x64.exe; its signature is checkedMicrosoft Software License Terms for the Visual C++ Redistributable (shown by Microsoft's installer); not redistributed by FoxySing

Models

Rows marked add yourself are not shipped and never downloaded by FoxySing; the others are downloaded by the installer or when first needed.

ModelUsed forAuthor / sourceLicense
BS-RoFormer SW BS-Roformer-SW (6 stems: voice, drums, bass, guitar, piano, other) — add yourselfseparation of voice, guitar and the restjarredou (the author's Hugging Face account no longer exists)not stated by the author (model card: "unknown"); some catalogs label it CC BY-NC-SA 4.0. No rights are granted here: use it for personal, non-commercial purposes only
Mel-Band RoFormer mel_band_roformer_karaoke_becruily — add yourselflead / backing vocals (the «Backing vocals» tick box)becruily, https://huggingface.co/becruily/mel-band-roformer-karaokefree for non-commercial use; commercial use by agreement with the author (author's statement in the model's discussion #1)
RMVPE rmvpe.onnxvocal pitch (notes)RVC project (lj1995/VoiceConversionWebUI), via the rmvpe-onnx package; original model Dream-High/RMVPE; FoxySing downloads the file itself from huggingface.co (or the mirror hf-mirror.com) and checks its SHA-256MIT (the RVC repository); the original model is Apache-2.0 — see «RMVPE: where the file comes from»
Dereverb Mel-Band RoFormer dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050 — add yourself (the «Dereverb» checkbox and the song button)echo removal from the voice before the pitch is foundanvuew, https://huggingface.co/anvuew/dereverb_mel_band_roformerGPL-3.0 (model card)
SwiftF0 swift-f0 0.3.0 with its built-in model (optional; the «SwiftF0» item in the installer, or on first use of «Better notes» — the «On» switch at SwiftF0 in ⚙ → «Models» — which improves the notes)second pitch model that refines RMVPELars Nieradzik, https://github.com/lars76/swift-f0 (PyPI)MIT
Mel-Band RoFormer Kim vocals_mel_band_roformer (ticked in the installer; chosen in ⚙ → «Models»)voice and music, no separate instrumentsKimberley Jensen, https://huggingface.co/KimberleyJSN/melbandroformer; FoxySing downloads the file from this page and checks its SHA-256MIT (model card, checked 18 Nov 2026)
Demucs v4 htdemucs_6s (files htdemucs_6s.yaml, 5c90dfd2-34c22ccb.th) — add yourselfvoice, drums, bass, guitar, piano, otherMeta (facebookresearch), https://github.com/facebookresearch/demucscode: MIT. The pretrained weights are not covered by the code license: the author (Alexandre Défossez) states they are provided for research purposes (demucs issues #267, #327, #508); the model card no longer states a license. Personal use only. Its quantization helper diffq (non-commercial license) is not installed: this model does not need it

Bundled with the page

ComponentUsed forLicense
static/vendor/SignalsmithStretch.mjs — Signalsmith Stretch Web Audio 1.3.2 (WASM/AudioWorklet, https://github.com/Signalsmith-Audio/signalsmith-stretch)live key shift and slow-down in the browserMIT (static/vendor/signalsmith-stretch.LICENSE and static/vendor/signalsmith-linear.LICENSE)
static/vendor/qrcode.mjs — QR Code Generator for JavaScript 1.4.4 by Kazuhiko Arase (https://github.com/kazuhikoarase/qrcode-generator; one export line added)the QR code of the «Sing together» invitation linkMIT (static/vendor/qrcode.LICENSE)

Python packages (exact versions from requirements.lock)

Notable licenses: soxr — LGPL-2.1-or-later; certifi, tqdm — MPL-2.0. The rest are permissive (MIT, BSD, Apache-2.0, ISC, PSF). audioop-lts and the standard-* packages bring back standard-library modules that Python 3.13 removed (needed by pydub, librosa and audioread); they are installed only on Python 3.13 and newer.

PackageVersionLicense
absl-py2.5.1Apache-2.0
annotated-doc0.0.5MIT
annotated-types0.8.0MIT
anyio4.16.0MIT
audio-separator0.48.0MIT
audioop-lts0.2.2PSF-2.0
audioread3.1.0MIT
beartype0.18.5MIT
beautifulsoup44.15.0MIT License
certifi2026.10.23MPL-2.0
cffi2.1.1MIT-0
charset-normalizer3.5.2MIT
click8.5.0BSD-3-Clause
cloudpickle3.1.2BSD-3-Clause
colorama0.4.6BSD License
cryptography50.0.2Apache-2.0 OR BSD-3-Clause
decorator5.3.1BSD-2-Clause
einops0.8.2MIT
fastapi0.162.0MIT
filelock4.2.6MIT
flatbuffers25.12.19Apache 2.0
fsspec2026.9.0BSD-3-Clause
h110.16.0MIT
hf-xet1.8.0Apache-2.0
httpcore22.16.0BSD-3-Clause
httpx22.16.0BSD-3-Clause
huggingface_hub2.4.0Apache-2.0
idna3.20BSD-3-Clause
Jinja23.1.6BSD License
joblib1.6.0BSD-3-Clause
julius0.2.8MIT License
lazy-loader0.6BSD-3-Clause
librosa0.11.0ISC
llvmlite0.50.0BSD-2-Clause AND Apache-2.0 WITH LLVM-exception
MarkupSafe3.0.4BSD-3-Clause
ml_collections1.1.0Apache Software License
ml_dtypes0.6.0Apache-2.0
mpmath1.3.0BSD
msgpack1.2.4Apache-2.0
narwhals2.28.0MIT
networkx3.7BSD-3-Clause
numba0.68.1BSD
numpy2.5.4BSD-3-Clause AND 0BSD AND MIT AND Zlib AND CC0-1.0
onnx-weekly1.24.0.dev20261102Apache-2.0
onnx2torch-py3131.6.0Apache License
onnxruntime1.31.1MIT License
opentelemetry-api1.45.2Apache-2.0
packaging26.3Apache-2.0 OR BSD-2-Clause
pillow13.0.0MIT-CMU
platformdirs4.14.3MIT
pooch1.9.0BSD-3-Clause
protobuf7.36.23-Clause BSD License
pycparser3.12BSD-3-Clause
pydantic2.14.2MIT
pydantic_core2.50.2MIT
pydub0.25.1MIT
python-multipart0.0.33Apache-2.0
PyYAML6.0.3MIT
RapidFuzz3.14.6MIT
requests2.34.2Apache-2.0
resampy0.4.3ISC
rmvpe-onnx0.2.3MIT
rotary-embedding-torch0.6.5MIT
samplerate0.1.0MIT
scikit-learn1.9.1BSD-3-Clause
scipy1.18.1BSD License
setuptools84.2.1MIT
six1.17.0MIT
soundfile0.14.0BSD 3-Clause License
soupsieve4.0.1MIT
soxr1.1.0LGPL-2.1-or-later
standard-aifc3.13.0PSF-2.0
standard-chunk3.13.0PSF-2.0
standard-sunau3.13.0PSF-2.0
starlette1.10.0BSD-3-Clause
sympy1.14.0BSD
syncedlyrics1.0.1MIT
threadpoolctl3.7.0BSD-3-Clause
tinytag2.3.2MIT
torch2.14.1BSD-3-Clause
torchvision0.29.2BSD
tqdm4.70.1MPL-2.0 AND MIT
truststore0.10.5MIT
typing-inspection0.4.4MIT
typing_extensions4.16.0PSF-2.0
urllib32.8.0MIT
uvicorn0.54.0BSD-3-Clause

The table is generated from the installed environment by a maintainer script (dev/licenses.py in the development repository).

Optional content sources and tools

FoxySing is only a client program for these: it does not own, run or control them, is not affiliated with them, and does not license the content they return. The licenses of the software components below do not grant any rights to the audio recordings, musical compositions or lyrics retrieved through them.

Source or toolWhat FoxySing does with itStatus
YouTubegets the audio of a link you paste, through yt-dlp (optional, experimental, personal use only)unofficial integration
yt-dlpdownloads and converts the available mediaUnlicense
yt-dlp-ejspart of yt-dlp (solves the JavaScript challenges of the page)Unlicense / MIT / ISC
Denoruns the JavaScript that yt-dlp needsMIT
LRCLIB (lrclib.net)gets lyrics and their timing (default)public API
lyrics.ovhgets lyrics (default)third-party service API
Genius, Musixmatch, NetEase, Megalobiz (via syncedlyrics)get lyrics («unofficial sources»: off by default, experimental, personal use only)unofficial sources
Paxsenix (Apple Music, Spotify, Musixmatch)gets lyrics («unofficial sources», only with your own Paxsenix key)third-party service

External services

ServiceUsed forNotes
LRCLIB (lrclib.net)lyrics (default)open API
lyrics.ovhlyrics (default)open API
Genius, Musixmatch, NetEase, Megalobiz, Apple Music, Spotify (via syncedlyrics and Paxsenix)lyrics (unofficial, off by default)not official APIs; may break or violate the services' terms
ElevenLabs (elevenlabs.io, Scribe)lyrics recognition by ear (optional; only on a button; your own key)the voice track of the song is sent to their servers; a paid service with its own limits and terms
MVSEP (mvsep.com)cloud separation (optional «Cloud» mode; your own key)the song is uploaded to their servers and processed there; uses your MVSEP credits; their terms apply
freqblog.comthe song's key (optional; your own key)the song's artist and title are sent; their terms apply
Cloudflare Quick Tunnel (trycloudflare.com)the «Sing together» link for the whole internet (optional; only while the guest entrance is open)guests' traffic goes through Cloudflare's servers; no account; Cloudflare offers quick tunnels for testing, without uptime guarantees; their terms apply
GitHubupdatesthe app checks this repository for new versions

Notices for other bundled or downloaded components

These components are used as they are, without changes. Their license texts and copyright notices are kept with the program files.