Skip to content

A PyTorch 1.0 implementation of the convolutions described in SincNet

License

Notifications You must be signed in to change notification settings

Popgun-Labs/SincNetConv

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SincConv

A PyTorch 1.0 implementation of the bandpass convolutions described in Interpretable Convolutional Filters with SincNet. Compared with normal convolution, this has the following practical benefits for audio-domain models such as speaker or phoneme recognition:

  • Fewer parameters
  • Faster convergence
  • Intepretable filters
  • Better performance

Adapted from the official implementation at: https://github.com/mravanelli/SincNet/. Compared to the original implementation, the filter bank construction has been parallelised. Additionally, padding has been added to preserve the length / time dimension of the input audio.

Authors

If you use this code or part of it, please cite the original paper authors!

Mirco Ravanelli, Yoshua Bengio, “Speaker Recognition from raw waveform with SincNet” Arxiv

About

A PyTorch 1.0 implementation of the convolutions described in SincNet

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •