rvc-next
Copyright (C) 2026 licyk and contributors. Licensed under the GNU General Public License v3.0
(see LICENSE). The application skeleton is reused from SD Model Hub and Hanaikada (GPLv3).

This product ports numerical code from the following projects, under their licences:

Retrieval-based-Voice-Conversion-WebUI (RVC)
  https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI
  MIT License. Copyright (c) 2023 liujing04, 源文雨, Ftps.
  Ported: the synthesizer and discriminator networks (rvc_next/engine/models), the offline and
  realtime conversion maths (rvc_next/engine/convert, rvc_next/engine/stream), RMVPE
  (rvc_next/engine/f0/rmvpe.py), TorchGate (rvc_next/engine/audio/denoise.py), the CUDA Graph
  cache (rvc_next/engine/graph.py), and the training stages (rvc_next/engine/train).

torchaudio (https://github.com/pytorch/audio)
  BSD 2-Clause License. Copyright (c) 2017 Facebook Inc. (Soumith Chintala). All rights reserved.
  Ported: the windowed-sinc resampler (``transforms.Resample`` and its kernel functions) into
  rvc_next/engine/audio/sinc_resample.py, which carries the licence text in full.

FCPE / torchfcpe (https://github.com/CNChTu/FCPE)
  MIT License. Copyright (c) 2023 CN_ChiTu. The conformer comes from Diffusion-SVC
  (https://github.com/CNChTu/Diffusion-SVC), MIT License.
  Ported: the bundled model's inference path (mel front end, convolution-only CFNaiveMelPE,
  decoders, loader) into rvc_next/engine/f0/fcpe_model.py. Its weights (fcpe_c_v001.pt) are
  downloaded as the "fcpe" asset.

Applio (https://github.com/IAHispano/Applio)
  MIT License. Copyright (c) 2026 AI Hispano.
  Ported: the RMVPE high-register corrector (rvc_next/engine/f0/high_register.py, contributed to
  Applio by dhayanithi-svaromance) and SwiftF0's subharmonic repair (rvc_next/engine/f0/swift.py).
  Downloaded from its repository (huggingface.co/IAHispano/Applio, MIT): the content-feature models
  SPIN, SPIN v2 and the Chinese (TencentGameMate, MIT), Japanese (rinna, Apache-2.0) and Korean
  (team-lucid, Apache-2.0) HuBERT models, in the transformers format Applio converted them to.
  The community base models are downloaded from their authors' own repositories, under whatever
  terms those state; rvc-next does not redistribute them.

torchcrepe (https://github.com/maxrmorrison/torchcrepe)
  MIT License. Copyright (c) 2020 Max Morrison. CREPE: https://github.com/marl/crepe, MIT License,
  Copyright (c) 2018 Jong Wook Kim.
  Ported: the CREPE network, framing, decoders and filters RVC uses (rvc_next/engine/f0/crepe_model.py).
  Its weights (full.pth, tiny.pth) are downloaded as the "crepe" and "crepe-tiny" assets.

SwiftF0 (https://github.com/lars76/swift-f0)
  MIT License. Copyright (c) 2025 Lars Nieradzik.
  Ported: the swift-f0 0.3.0 ONNX graph as a PyTorch module (rvc_next/engine/f0/swift_model.py);
  its weights ship as rvc_next/engine/f0/data/swift_f0.pt (scripts/extract_swift_f0.py).

PyMSS (music source separation), used as a dependency, MIT License.
edge-tts (https://github.com/rany2/edge-tts), the optional text to speech, used as a dependency, LGPL-3.0.
pedalboard (https://github.com/spotify/pedalboard), the optional effects chain, used as a dependency,
  GPL-3.0. Copyright (c) 2021-2023 Spotify AB.

RMVPE: https://github.com/Dream-High/RMVPE, trained by yxlllc and RVC-Boss.
DDSP-SVC (the SOLA crossfade): https://github.com/yxlllc/DDSP-SVC, MIT License.
TorchGate (spectral gating), by Asaf Zorea, as vendored in RVC (noisereduce, MIT License).

Model weights downloaded at run time (HuBERT/ContentVec, RMVPE, the pretrained base models and
the separation checkpoints) come from https://huggingface.co/lj1995/VoiceConversionWebUI and
carry their own terms.
