Skip to content

research on convolutional neural networks in fourier space

Notifications You must be signed in to change notification settings

rileyedmunds/complexcnn

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Complex Convolutional Neural Networks for Environmental Sound Classification

[Prepublication] Abstract: In this paper we introduce a new framework and approach for convolutional neural network computation. By introducing layer functions that intelligently process complex-domain data in deep neural network architectures, we improve upon prior understanding and performance in complex-valued convolutional neural networks. Using novel derivations of convolutional, down-sampling, non-linear, and affine layers implemented in a complex-valued counterpart to Caffe, we proved results when evaluated against real-valued models in the context of environmental sound classification. Finally, we demonstrated these results are applicable to many other fields, including, but not limited to MRI imaging, signal processing, and LiDAR mapping.

Questions? Contact edmunds@ml.berkeley.edu. More information on my website.

Credits: the real-valued cNN model is based on Karol J. Piczak's paper, Environmental Sound Classification with Convolutional Neural Networks.

License

MIT License

Copyright (c) 2016-2017 Riley F. Edmunds

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.


ESC-50: Dataset for Environmental Sound Classification

Download | A peek inside | License

The ESC-50 dataset is a public labeled set of 2000 environmental recordings (50 classes, 40 clips per class, 5 seconds per clip) suitable for environmental sound classification tasks.

See ESC: Dataset for Environmental Sound Classification - paper replication data for the full paper with a more thorough analysis.

The dataset consists of 50 classes of recordings in 5 loosely defined groups:

Animals

Natural soundscapes & water sounds

Human, non-speech sounds

Interior/domestic sounds

Exterior/urban noises

Clips have been constructed from public field recordings gathered by the Freesound.org project. The dataset has been prearranged into 5 folds. Clips stemming from the same original source file are contained in a single fold.

File naming scheme:

category_id - category_name/fold_number-Freesound_clip_ID-take_letter.ogg

File details:

5-second-long recordings reconverted to a unified format:
- 44100 Hz,
- single channel (monophonic),
- Vorbis/Ogg compression @ 192 kbit/s. 

Download

The dataset can be downloaded as a single .zip file (~200 MB):

ESC-50 dataset

A peek inside

Waveforms and mel-spectrograms of ESC-50 dataset recordings:

Waveforms and mel-spectrograms of ESC-50 dataset recordings

License

The dataset is available under the terms of the Creative Commons license - Attribution-NonCommercial.

A smaller subset (ESC-10) is available under CC BY (Attribution).

In academic settings please cite:

K. J. Piczak. ESC: Dataset for Environmental Sound Classification. In Proceedings of the ACM International Conference on Multimedia, in press, ACM, 2015.

[DOI: http://dx.doi.org/10.1145/2733373.2806390]

Due to GitHub limitations (README length limit) licensing details for individual clips are available in the plain text README.



About

research on convolutional neural networks in fourier space

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages