97 lines
2.7 KiB
Plaintext
Executable File
97 lines
2.7 KiB
Plaintext
Executable File
# Web framework
|
|
fastapi==0.115.6
|
|
uvicorn[standard]==0.34.0
|
|
python-multipart==0.0.20
|
|
aiofiles==24.1.0
|
|
httpx>=0.27.0,<1.0
|
|
|
|
# Database
|
|
sqlalchemy==2.0.36
|
|
|
|
# Audio utilities
|
|
pydub==0.25.1
|
|
soundfile>=0.13.1
|
|
librosa==0.10.2
|
|
mutagen==1.47.0
|
|
Pillow>=10.4.0
|
|
|
|
# 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>=2.1.0
|
|
|
|
# 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>=0.8.1
|
|
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).
|
|
# Pin below 1.27 because newer wheels in this stack require libcudart.so.13.
|
|
onnxruntime-gpu==1.20.1
|
|
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
|
|
|
|
# yt-dlp import integration
|
|
yt-dlp>=2026.01.01
|
|
|
|
# MIDI transcription via basic-pitch
|
|
# Installed separately in Dockerfile with --no-deps to avoid TF/numpy conflict.
|
|
# Runtime inference uses onnxruntime (above) instead of TensorFlow.
|
|
# pretty_midi and mir_eval are basic-pitch deps that must be installed separately
|
|
# since --no-deps skips them along with tensorflow.
|
|
pretty_midi>=0.2.10
|
|
mir_eval>=0.7
|
|
|
|
# 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
|
|
accelerate>=0.19.0
|
|
|
|
# audio-separator (required by SCNet, MedleyVox, Apollo, Aspiration, Bandit separators)
|
|
audio-separator[gpu]>=0.28.0
|
|
|
|
# Lyric transcription via WhisperX (faster-whisper backend + phoneme-level word alignment)
|
|
whisperx>=3.8.1
|
|
|
|
# Karaoke-oriented lyric line segmentation
|
|
karaoke-lyrics-processor>=0.6.0
|
|
|
|
# Phoneme-level syllable alignment (requires espeak-ng system package)
|
|
# Used by the "Phoneme alignment" option in AI Transcribe to produce
|
|
# syllable-level ELRC timestamps via facebook/wav2vec2-xlsr-53-espeak-cv-ft.
|
|
phonemizer>=3.2.0
|
|
|
|
# Karaoke gloss transliteration/phonetics engines
|
|
nltk>=3.9.1
|
|
pypinyin>=0.53.0
|
|
pykakasi>=2.3.0
|
|
korean-romanizer>=0.28.0
|
|
indic-transliteration>=2.3.60
|
|
transliterate>=1.10.2
|
|
eng-to-ipa>=0.0.2
|
|
pythainlp>=5.0.4
|