Spherical room impulse response dataset generation for AmbiSep

This toolbox can be used to generate simulated spherical room impulse response (SRIR) datasets.

Prerequisites

Python and Matlab, a C/C++ compiler

Installation

Git

git clone --recurse-submodules <repo-URL>

Python

• The following Python packages are required
  • pyyaml
  • soundfile
  • numpy
  • scipy
  • pandas
• Create your environment (e.g. via conda create -n <envname> or via python -m venv .venv)
• Activate your environment (e.g. via conda activate <envname> or via .venv/Scripts/activate)
• You can install the required packages via pip install -r requirements.txt

SMIR generator MEX

1. Open Matlab
2. Go to ./matlab/SMIR-Generator
3. Check that a compiler is set for mex via mex -setup
4. Run (in Matlab): mex smir_generator_loop.cpp

You can find more information about the SMIR generator here.

Generating the AmbiSep SRIR dataset

Warning: This will take a long time! You can change ANECHOIC to true in configs/srir_config.yml for quick checking that the pipeline is working.

Run generate_srirs.sh or generate_srirs_parallel.sh for faster generation on multicore CPUs. In the latter case, you can adjust the number of CPU cores used in the shell script. The output SRIRs will be in srirs/medt60_set_05t. The mic folder contains simulated spherical array impulse responses. To convert these to Ambisonics (First order AmbiX format), run python srirs_to_ambix.py in your Python environment.

Generate a different SRIR dataset

Either adjust the parameters in srir_config.yml or in srir_config.py. In the latter case, you have to run srir_config.py to update the YAML file.