# Core dependencies for Step Audio EditX TTS Node
# Install with: pip install -r requirements.txt

# Audio processing
librosa>=0.10.2
# CRITICAL: soundfile is required for containerized/headless environments (RunPod, Docker, etc.)
# Without soundfile, torchaudio falls back to torchcodec which doesn't support BytesIO properly
soundfile>=0.12.1
torchaudio>=2.0.0
onnxruntime>=1.17.0

# Model loading and inference
# CRITICAL: transformers must be exactly 4.53.3 for correct token generation
# Newer versions (4.54+) have a tokenization bug that causes silent audio
transformers==4.53.3
accelerate>=0.20.0
sentencepiece>=0.1.99

# Utilities
omegaconf>=2.3.0
hyperpyyaml>=1.2.2
protobuf>=3.20.0
gradio==5.49.1

# Whisper (required by tokenizer and frontend)
openai-whisper>=20231117

# Optional: Quantization support (uncomment if needed)
# autoawq>=0.2.0
# bitsandbytes>=0.41.0

# Note: numpy, torch, pillow should already be installed by ComfyUI
# funasr is NOT required - we use bundled funasr_detach instead
# torchcodec is NOT required - soundfile handles all audio I/O
