AUV

Teaching Audio Universal Vector Quantization with Single Nested Codebook

Setup

pip install auv
wget https://huggingface.co/SWivid/AUV/resolve/main/auv.pt

Inference

Command line usage, reconstruct all .wav files under the input-dir and write to the output-dir:

auv-infer --input-dir INPUT_WAV_DIR --output-dir OUTPUT_WAV_DIR --ckpt CKPT_PATH
# if torch.bfloat16 inference: --bf16
# if need to assign gpu: --device cuda:0

Python script usage see src/auv/infer.py.