This repository contains the official PyTorch implementations for the paper:
- Danilo de Oliveira, Julius Richter, Jean-Marie Lemercier, Simon Welker, Timo Gerkmann, "Non-intrusive Speech Quality Assessment with Diffusion Models Trained on Clean Speech", accepted at ISCA Interspecch 2025. [bibtex]
The code is largely based on the repository of [1]
- Create a new virtual environment with Python >= 3.11 (we have not tested other Python versions, but they may work).
- Install the package dependencies via
pip install -r requirements.txtand let pip resolve the dependencies for you - Install the EDM2 code as a submodule:
git submodule update --init --recursive
For training, we use the EARS-WHAM dataset.
- The checkpoint used in the paper can be downloaded here
Training is done by executing train_sqa.py. A minimal running example with default settings (as in our paper) can be run with
torchrun --standalone --nproc_per_node=<num-gpus> train_sqa.py --outdir=<log-dir> --data=<path-to-trainset> --batch-gpu=<batch-size-per-gpu>where <path-to-trainset> should be a path to a folder containing clean .wav files (subdirectories are also supported).
To reconstruct a new EMA profile with length 0.08, run
python edm2/reconstruct_phema.py --indir=<log-dir> --outdir=<reconstructed-ema-dir> --outstd=0.080For more detailed on post-hoc EMA reconstruction, please refer to the EDM2 repository.
To calculate the diffusion log likelihoods on a test set and save them in a csv file, run
torchrun --standalone --nproc_per_node=<num-gpus> calculate_likelihood.py --checkpoint=<path-to-pkl> --data_dir=<path-to-testset> --output_file=<path-to-csv>The --checkpoint parameter should be the path to a snapshot or a reconstructed EMA profile.
The code and checkpoints are licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
We kindly ask you to cite our papers in your publication when using any of our research or code:
@misc{deoliveira2024nonintrusivespeechqualityassessment,
title={Non-intrusive Speech Quality Assessment with Diffusion Models Trained on Clean Speech},
author={Danilo de Oliveira and Julius Richter and Jean-Marie Lemercier and Simon Welker and Timo Gerkmann},
year={2024},
eprint={2410.17834},
archivePrefix={arXiv},
primaryClass={eess.AS},
url={https://arxiv.org/abs/2410.17834},
}[1] Tero Karras, Miika Aittala, Jaakko Lehtinen, Janne Hellsten, Timo Aila, Samuli Laine, "Analyzing and Improving the Training Dynamics of Diffusion Models", CVPR 2024. [Code]