Skip to content

Latest commit

 

History

History
86 lines (74 loc) · 3.94 KB

README.md

File metadata and controls

86 lines (74 loc) · 3.94 KB

PubSep

PWC

This repository provides training and evalution scripts for the DNN speech separation models described in various papers

A baseline TCN model (from SpeechBrain) is also provided along with tools for estimating computational efficiency.

This recipe is a fork of the WHAMandWHAMR recipe in the SpeechBrain library (required, see below). For more help and information on any SpeechBrain related issues:

Data and models

Data:

  • WHAMR
  • WSJ0-2Mix
  • LibriMix [WIP]

Models:

  • Time-Domain Conformers (TD-Conformer)
  • Deformable Temporal Convolutional Networks (DTCN)
  • Utterance-Weighted Multi-Dilation Temporal Convolutional Network (WD-TCN)
  • Temporal Convolutional Networks (Conv-TasNet without skip connections)

Running basic script

First install SRMRpy and remaining required packages

git clone /~https://github.com/jfsantos/SRMRpy.git
cd SRMRpy
python setup.py install

pip install -r requirements.txt

Then to run basic training of a DTCN model firstly change the data_folder hyperparameter in the separation/hparams/deformable/dtcn-whamr.yaml folder. Then run

cd separation
HPARAMS=hparams/deformable/dtcn-whamr.yaml
python train.py $HPARAMS

or if you wish to use multi GPU (recommended) run

python -m torch.distributed.launch --nproc_per_node=$NGPU train.py $HPARAMS --distributed_launch --distributed_backend='nccl'

replacing NGPU with the desired number of GPUs to use. In order to use dynamic mixing you will also need to change the base_folder_dm and rir_path hyperparameters, refer to /~https://github.com/speechbrain/speechbrain/blob/develop/recipes/WHAMandWHAMR/separation/README.md for more info on setting up dynamic mixing in SpeechBrain recipes.

Known issues

  • The main issue at present is mixed precision training for DTCN with autocast enabled. We do not recommend trying to use this functionality at present.

Paper

Please cite the following papers if you make use of the respective part of this codebase:

@INPROCEEDINGS{tdconformer23,
  author={Ravenscroft, William and Goetze, Stefan and Hain, Thomas},
  booktitle={2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU)}, 
  title={On Time Domain Conformer Models for Monaural Speech Separation in Noisy Reverberant Acoustic Environments}, 
  year={2023},
  volume={},
  number={},
  pages={1-7},
  doi={10.1109/ASRU57964.2023.10389669}}

@INPROCEEDINGS{dtcn23,
  author={Ravenscroft, William and Goetze, Stefan and Hain, Thomas},
  booktitle={ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  title={Deformable Temporal Convolutional Networks for Monaural Noisy Reverberant Speech Separation},
  month={June}
  year={2023},
  volume={},
  number={},
  pages={1-5},
  doi={10.1109/ICASSP49357.2023.10095230}}

@INPROCEEDINGS{wdtcn22,
  author={Ravenscroft, William and Goetze, Stefan and Hain, Thomas},
  booktitle={2022 International Workshop on Acoustic Signal Enhancement (IWAENC)}, 
  title={Utterance Weighted Multi-Dilation Temporal Convolutional Networks for Monaural Speech Dereverberation}, 
  year={2022},
  volume={},
  number={},
  pages={1-5},
  doi={10.1109/IWAENC53105.2022.9914752}}