This repository contains the official PyTorch implementations for the 2023 paper:
- Reducing the Prior Mismatch of Stochastic Differential Equations for Diffusion-based Speech Enhancement [1]
This repository builds upon our previous work, that can be found here https://github.com/sp-uhh/sgmse
- Create a new virtual environment with Python 3.8 (we have not tested other Python versions, but they may work).
- Install the package dependencies via
pip install -r requirements.txt
. - If using W&B logging (default):
- Set up a wandb.ai account
- Log in via
wandb login
before running our code.
- If not using W&B logging:
- Pass the option
--no_wandb
totrain.py
. - Your logs will be stored as local TensorBoard logs. Run
tensorboard --logdir logs/
to see them.
- Pass the option
Training is done by executing train.py
. A minimal running example with default settings (as in our paper [1]) can be run with
python train.py --base_dir <your_base_dir>
where your_base_dir
should be a path to a folder containing subdirectories train/
and valid/
(optionally test/
as well). Each subdirectory must itself have two subdirectories clean/
and noisy/
, with the same filenames present in both. We currently only support training with .wav
files. To reproduce results in [1] you could use the following training settings on the wsj0-chime3 dataset:
python train.py --base_dir <your_base_dir> --batch_size 16 --backbone ncsnpp --sde bbed --t_eps 0.03 --gpus 1 --num_eval_files 10 --spec_abs_exponent 0.5 --spec_factor 0.15 --loss_abs_exponent 1 --loss_type mse --k 2.6 --theta 0.51
To get the training set, we refer to https://github.com/sp-uhh/sgmse and execute create_wsj0_chime3.py.
To see all available training options, run python train.py --help
. Note that the available options for the SDE and the backbone network change depending on which SDE and backbone you use. These can be set through the --sde
and --backbone
options.
To evaluate on a test set, run
python eval.py --test_dir <your_test_dir> --folder_destination <your_enhanced_dir> --ckpt <path_to_model_checkpoint>
to generate the enhanced .wav files. For instance,
python eval.py --test_dir <your_test_dir> --folder_destination <your_enhanced_dir> --ckpt <path_to_model_checkpoint> --N 30 --reverse_starting_point 0.5 --force_N 15
starts enhancement from 0.5 with 15 reverse steps. This would be the result of Tab. 1 last row in [1], when the provided checkpoint (download from here https://drive.google.com/file/d/1_h7pH6o-j7GV_E69SbRQF2BMRlC8tmz_/view?usp=share_link) is loaded in the checkpoint folder. This is the checkpoint that was used to produce the results in [1].
We kindly ask you to cite our paper (can be found on https://arxiv.org/abs/2302.14748) in your publication when using any of our research or code:
[1] Bunlong Lay, Simon Welker, Julius Richter and Timo Gerkmann. Reducing the Prior Mismatch of Stochastic Differential Equations for Diffusion-based Speech Enhancement, ISCA Interspeech, 2023.