Inference and Denoise: Causal Inference-based Neural Speech Enhancement

This is the official implementation of "Inference and Denoise: Causal Inference-based Neural Speech Enhancement".
We are updating our code and additional supplementary material and soon. Thank you.

Requirements

• pytorch 1.12.1
• pytorch_lightning 1.7.5
• SOTA SE models (slightly modified to fit the usage of CISE) are credited to:
  • MANNER
  • CMGAN
  • DEMUCS

Usage

To train and inference, please run

python main.py \
  --save_path <path/to/checkpoint/and/tensorboard/> \
  --output_dir <path/to/prediction> \
  --use_pretrained \
  --batch_size 8 \
  --hop_length 320