vad_demo / requirements.txt
Gabriel Bibb贸
GitHub-faithful implementation - 32kHz, 2048 FFT, per-model delays, 80ms gaps
08ba0e7
# STABLE GRADIO 4.x VERSION - FIXES JSON SCHEMA BUG
gradio==4.42.0
# Core dependencies - compatible with Gradio 4.42.0
numpy>=1.24.0,<2.0.0
torch>=2.1.0,<2.4.0
torchaudio>=2.1.0,<2.4.0
# Audio processing - stable versions
librosa>=0.10.1,<0.11.0
soundfile>=0.12.1
scipy>=1.10.0,<1.14.0
# Visualization - compatible with Gradio 4.x
plotly>=5.15.0,<5.18.0
# ML libraries - Gradio 4.x tested versions
transformers>=4.30.0,<4.40.0
datasets>=2.14.0,<2.18.0
# PANNs inference tool - for easy PANNs model loading
panns-inference>=0.1.0
# AST and transformers dependencies
accelerate>=0.20.0
safetensors>=0.3.0
tokenizers>=0.13.0
# Optional dependencies with fallbacks
webrtcvad>=2.0.10; python_version >= "3.8" and sys_platform != "darwin"
scikit-learn>=1.3.0,<1.4.0
psutil>=5.9.0
# System utilities
matplotlib>=3.6.0,<3.8.0
# Pin pydantic to avoid conflicts (reported fix)
pydantic>=2.5.0,<2.8.0
# Additional dependencies for audio processing
resampy>=0.4.0
numba>=0.56.0
# For model downloads and caching
requests>=2.25.0
tqdm>=4.64.0
# Additional transformers ecosystem
huggingface-hub>=0.15.0