Skip to content

Latest commit

 

History

History
69 lines (57 loc) · 2.69 KB

File metadata and controls

69 lines (57 loc) · 2.69 KB

Third-party notices

ApexSVC's own code is licensed under AGPL-3.0-or-later (see LICENSE). The components listed below are vendored, wrapped, or pulled at runtime. Each keeps its original license and authorship; the SPDX identifier next to each entry is the upstream license, not ApexSVC's. Visit the linked upstreams for full license texts.

NeuCoSVC / NeuCoSVC2

  • Upstream: https://github.com/thuhcsi/NeuCoSVC (branch NeuCoSVC2)
  • Paper: Sha et al., "Neural Concatenative Singing Voice Conversion", arXiv:2312.04919
  • License: MIT
  • Files vendored: the NSF HiFi-GAN generator (GeneratorNSF, source-filter excited HiFi-GAN), kNN matcher, and small audio utility helpers, under svc/synth/, svc/match/, and svc/utils/. The training code, dataset scaffolding, REAPER pitch wrapper, and Phoneme Hallucinator are NOT vendored.

WavLM (Microsoft)

TorchFCPE

praat-parselmouth (optional F0 backend)

librosa (optional F0 backend)

kNN-VC

Silero VAD

PyTorch / torchaudio

  • Upstream: https://pytorch.org
  • License: BSD-3-Clause
  • Pinned to the cu121 wheel index in pyproject.toml for Windows CUDA support.

Gradio

Other runtime dependencies

Standard scientific Python stack (NumPy, SciPy, soundfile, resampy, tqdm, click, Hugging Face huggingface_hub / transformers / safetensors / omegaconf, gdown). All used as regular pip packages under their respective permissive licenses (BSD, MIT, Apache-2.0). See pyproject.toml for the full pinned list.