This is the code for our paper RIR-in-a-Box: Estimating Room Acoustics from 3D Mesh Data through Shoebox Approximation, accepted at Interspeech 2024.
This setup tutorial is functional, but work is still being done to simplify the reuse process.
Install the cuda 12.1 toolkit. Other versions may work, but this is the one I used. If you use another other, please change the pytorch-cuda version in the conda install command below.
conda create -n rirbox python=3.9
conda activate rirbox
conda install numpy
conda install pytorch torchvision torchaudio pyg pytorch-cuda=12.1 -c pytorch -c nvidia -c pyg
python -m pip install python-dateutil
python -m pip install soundfile
python -m pip install pandas
python -m pip install scipy
python -m pip install librosa
python -m pip install easydict
python -m pip install cupy-cuda12x
python -m pip install wavefile
python -m pip install torchfile
python -m pip install pyyaml==5.4.1
python -m pip install pymeshlab
python -m pip install openmesh
python -m pip install gdown
python -m pip install matplotlib
python -m pip install IPython
python -m pip install pydub
python -m pip install auraloss
python -m pip install wandb
python -m pip install pyroomacoustics
python -m pip install fake-bpy-module
python -m pip install trimesh
python -m pip install tqdm
python -m pip install libigl
python -m pip install shapely
python -m pip install pyvista
conda install pytorch-scatter -c pyg -c nvidia
Go to the RIR-in-a-Box/datasets/GWA_3DFRONT folder for extra instructions.
Then, go to the RIR-in-a-Box/models folder for extra instructions.
Congrats! Enjoy RIR-in-a-Box.
conda activate rirbox
To train models. You can choose your own config files by modifying the script.
python run_trainings.py
To validate model performance / view validation results. You can choose your own config files by modifying the script. You can also choose which validation experiments to run / show by modifying the script.
python run_validation.py
To monitor nvidia gpu usage in another terminal.
watch -d -n 0.5 nvidia-smi
If you find our work or code repository useful in your research, please cite our paper:
Liam Kelley, Diego Di Carlo, Aditya Arie Nugraha, Mathieu Fontaine, Yoshiaki Bando, et al.. RIR-in-a-Box: Estimating Room Acoustics from 3D Mesh Data through Shoebox Approximation. INTERSPEECH, Sep 2024, Kos International Convention Center, Kos Island, Greece. ⟨hal-04632526⟩
Thank you for your support!