This repository contains supplementary code for the paper "Imposing Correlation Structures for Deep Binaural Spatio-Temporal Wiener Filtering" by M. Tammen and S. Doclo, IEEE Trans. Audio, Speech and Language Processing, vol. 33, pp. 1278-1292, 2025.
- Clone this repository to your machine.
- Make sure Anaconda or Miniconda are available.
- Create and activate the conda environment using the provided
environment.ymlfile. We recommend using Mamba for a faster installation.# If you don't have mamba, install it first: # conda install -n base -c conda-forge mamba mamba env create -f environment.yml conda activate j2
You can use the inference.py script to enhance a noisy audio file using one of the pretrained models mentioned in the paper.
To run the inference script, use a command like the following:
python inference.py --model stwf_noCommonSTCM_noRTF --input data/noisy.wav --output data/noisy_enhanced.wavThis command will:
- Load the
stwf_noCommonSTCM_noRTFmodel. - Process the
data/noisy.wavfile. - Save the enhanced audio to
data/noisy_enhanced.wav.
If you don't specify an output file, the enhanced audio will be saved in the same directory as the input file with _enhanced appended to the name.
The following models are available for use with the --model argument (see Table II in the paper):
stwf_noCommonSTCM_noRTFstwf_CommonSTCM_noRTFstwf_CommonSTCM_globalRTFstwf_CommonSTCM_ipsiRTFstwf_bilat_CommonSTCM_noRTFstwf_bilat_CommonSTCM_globaldf_noRTF