Skip to content

harveyf2801/DNNAutoAlign

Repository files navigation

Differentiable All-Pass DNNs

This project aims to align an input signal with a target / reference signal using differentiable all-pass filters. The Parameter Network will control the individual parameters to align the phase of both signals. The filters consist of 2nd order all-pass filters, and are applied using the frequency sampling method to approximate a cascade of IIR filters. This method along with the main network architecture was adapted from the DASP library. A Temporal Convolutional Network (TCN) is used to train the model as it allows the model to learn based on the whole time domain signal and avoids phase estimation.

Citations

Differentiable All-Pass Filters

@inproceedings{bargum2023,
author = {Bargum, Anders and Serafin, Stefania and Erkut, Cumhur and Parker, Julian},
year = {2023},
month = {06},
pages = {},
title = {Differentiable Allpass Filters for Phase Response Estimation and Automatic Signal Alignment}
}

MR-STFT Loss

@inproceedings{steinmetz2020auraloss,
    title={auraloss: {A}udio focused loss functions in {PyTorch}},
    author={Steinmetz, Christian J. and Reiss, Joshua D.},
    booktitle={Digital Music Research Network One-day Workshop (DMRN+15)},
    year={2020}
}
@article{Arik_2019,
   title={Fast Spectrogram Inversion Using Multi-Head Convolutional Neural Networks},
   volume={26},
   ISSN={1558-2361},
   url={http://dx.doi.org/10.1109/LSP.2018.2880284},
   DOI={10.1109/lsp.2018.2880284},
   number={1},
   journal={IEEE Signal Processing Letters},
   publisher={Institute of Electrical and Electronics Engineers (IEEE)},
   author={Arik, Sercan O. and Jun, Heewoo and Diamos, Gregory},
   year={2019},
   month=jan, pages={94–98} }

Differentiable parametric EQ and dynamic range compressor

@article{steinmetz2022style,
  title={Style transfer of audio effects with differentiable signal processing},
  author={Steinmetz, Christian J and Bryan, Nicholas J and Reiss, Joshua D},
  journal={arXiv preprint arXiv:2207.08759},
  year={2022}
}

Differentiable IIR filters

@inproceedings{nercessian2020neural,
  title={Neural parametric equalizer matching using differentiable biquads},
  author={Nercessian, Shahan},
  booktitle={DAFx},
  year={2020}
}
@inproceedings{colonel2022direct,
  title={Direct design of biquad filter cascades with deep learning 
          by sampling random polynomials},
  author={Colonel, Joseph T and Steinmetz, Christian J and 
          Michelen, Marcus and Reiss, Joshua D},
  booktitle={ICASSP},
  year={2022},
  organization={IEEE}
}

About

DNN Auto Align | Automatically phase aligns a signal based off of the temporal features present in the time domain signal passed directly into the Temporal Convolution Network (TCN). Using MR-STFT loss to calculate the magnitude difference between the theoretical 'perfectly aligned' signal and the current state.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors