74 lines
1.9 KiB
Plaintext
Executable File
74 lines
1.9 KiB
Plaintext
Executable File
# Web framework
|
|
fastapi==0.115.6
|
|
uvicorn[standard]==0.34.0
|
|
python-multipart==0.0.20
|
|
aiofiles==24.1.0
|
|
|
|
# Database
|
|
sqlalchemy==2.0.36
|
|
|
|
# Audio utilities
|
|
pydub==0.25.1
|
|
soundfile==0.12.1
|
|
librosa==0.10.2
|
|
mutagen==1.47.0
|
|
|
|
# ABC notation
|
|
music21==9.3.0
|
|
|
|
# ML — PyTorch (CUDA 12.1 build; installed via --index-url in Dockerfile)
|
|
# torch and torchaudio are installed separately in the Dockerfile with the CUDA wheel index.
|
|
numpy==1.26.4
|
|
|
|
# Demucs stem separation
|
|
demucs==4.0.1
|
|
|
|
# BS-RoFormer model architecture dependencies (includes MelBandRoformer)
|
|
bs-roformer>=1.0.0
|
|
omegaconf==2.3.0
|
|
pyyaml>=6.0
|
|
einops==0.8.0
|
|
rotary-embedding-torch==0.6.5
|
|
beartype==0.18.5
|
|
tqdm>=4.66
|
|
|
|
# MelBand RoFormer dependencies
|
|
hyper-connections>=0.0.10
|
|
|
|
# MDX-Net dependencies (ONNX models from UVR)
|
|
# onnxruntime-gpu provides CUDAExecutionProvider so both MDX-Net and basic-pitch
|
|
# automatically run on the GPU when available (conflicts with onnxruntime CPU package).
|
|
onnxruntime-gpu>=1.16.0
|
|
onnx>=1.14.0
|
|
|
|
# HuggingFace CLI (for BS-RoFormer model download)
|
|
huggingface-hub[cli]>=0.24
|
|
|
|
# Subsonic server integration
|
|
requests>=2.31.0
|
|
cryptography>=42.0
|
|
|
|
# MIDI transcription via basic-pitch
|
|
basic-pitch>=0.3.0
|
|
|
|
# Syllable hyphenation (pure-Python, for per-syllable ELRC timing)
|
|
pyphen>=0.14.0
|
|
|
|
# Lyrics-sync dependencies (for ELRC alignment)
|
|
# Pin to <4.46 — newer releases change internal lazy-module wiring in ways that
|
|
# break Wav2Vec2ForCTC import when audio-separator also pins transformers deps.
|
|
transformers>=4.29.0,<4.46.0
|
|
accelerate>=0.19.0
|
|
|
|
# audio-separator (required by SCNet, MedleyVox, Apollo, Aspiration, Bandit separators)
|
|
audio-separator[gpu]>=0.28.0
|
|
|
|
# Lyric transcription via OpenAI Whisper (local, offline)
|
|
# Provides word-level timestamps for lyrics transcription.
|
|
# Models are downloaded automatically on first use.
|
|
openai-whisper>=20231117
|
|
|
|
# Optional: WhisperX for faster transcription + phoneme-level word alignment
|
|
# Uncomment to enable the WhisperX engine option in AI Transcribe:
|
|
whisperx>=3.1.0
|