Skip to content

Latest commit

 

History

History
67 lines (59 loc) · 3.07 KB

README.md

File metadata and controls

67 lines (59 loc) · 3.07 KB
Raw

Adbout the DNS Challenge

The Deep Noise Suppression (DNS) Challenge is a single-channel speech enhancement challenge organized by Microsoft, with a focus on real-time applications. More info can be found on the official page.

- Our basic recipe

This is made to make your life simpler and research easier !

What we automate for you :
  • Install git-lfs without root (required to download the data).
  • Download the data from the official repo.
  • Create the dataset with default parameters.
  • Ready-to-use DataLoader to train your net with.
  • Example scripts with all the ingredients for a successful system
  • MutliGPU support / Logging (+ Tensorboard) / LR scheduler (thanks lightning!)
What you can focus on :
  • Some new architectures to outperform our model
  • Fancy loss functions to improve speech quality
  • All the research, all the fun !
How to use it?
  • Need to install a python environment? Check this out !
  • Open run.sh, change storage_dir to a path where you can afford storing 320GB of data.
  • Just ./run.sh and it's on.
Some notes:
  • After the first execution, you can go and change stage=4 in run.sh to avoid redoing all the steps everytime.
  • To use GPUs for training, run run.sh --id 0,1 where 0 and 1 are the GPUs you want to use, training will automatically take advantage of both GPUs.
  • By default, a random id is generated for each run, you can also add a tag to name the experiments how you want. For example run.sh --tag with_cool_loss will save all results to exp/train_dns_with_cool_loss. You'll also find the corresponding log file in logs/train_dns_with_cool_loss.log.
  • If you want to change the data generation config, go the storage_dir, change the noisyspeech_synthesizer.cfg accordingly and restart from stage 2. Be aware that this will overwrite the previous json files in data/.

The data download, dataset creation and preprocessing will take a while (around a day in my case). From stage 4 (training), be sure you have enough compute power to train your DNN. Before that, you're I/O bound so not much compute power is needed.

References

  • The challenge paper, here.
@misc{DNSChallenge2020,
title={The INTERSPEECH 2020 Deep Noise Suppression Challenge: Datasets, Subjective Speech Quality and Testing Framework},
author={Chandan K. A. Reddy and Ebrahim Beyrami and Harishchandra Dubey and Vishak Gopal and Roger Cheng and Ross Cutler and Sergiy Matusevych and Robert Aichner and Ashkan Aazami and Sebastian Braun and Puneet Rana and Sriram Srinivasan and Johannes Gehrke}, year={2020},
eprint={2001.08662},
}
  • The baseline paper, here.
@misc{xia2020weighted,
title={Weighted Speech Distortion Losses for Neural-network-based Real-time Speech Enhancement},
author={Yangyang Xia and Sebastian Braun and Chandan K. A. Reddy and Harishchandra Dubey and Ross Cutler and Ivan Tashev},
year={2020},
eprint={2001.10601},
}