Skip to content

KinWaiCheuk/nnAudio

master
Switch branches/tags
Code

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 
 
 

nnAudio

nnAudio is an audio processing toolbox using PyTorch convolutional neural network as its backend. By doing so, spectrograms can be generated from audio on-the-fly during neural network training and the Fourier kernels (e.g. or CQT kernels) can be trained. Kapre has a similar concept in which they also use 1D convolutional neural network to extract spectrograms based on Keras.

Other GPU audio processing tools are torchaudio and tf.signal. But they are not using the neural network approach, and hence the Fourier basis can not be trained. As of PyTorch 1.6.0, torchaudio is still very difficult to install under the Windows environment due to sox. nnAudio is a more compatible audio processing tool across different operating systems since it relies mostly on PyTorch convolutional neural network. The name of nnAudio comes from torch.nn

Installation

pip install git+https://github.com/KinWaiCheuk/nnAudio.git#subdirectory=Installation

or

pip install nnAudio==0.3.1

Documentation

https://kinwaicheuk.github.io/nnAudio/index.html

Comparison with other libraries

Feature nnAudio torch.stft kapre torchaudio tf.signal torch-stft librosa
Trainable
Differentiable
Linear frequency STFT
Logarithmic frequency STFT
Inverse STFT
Griffin-Lim
Mel
MFCC
CQT
VQT
Gammatone
CFP1
GPU support

: Fully support ☑️: Developing (only available in dev version) : Not support

1 Combining Spectral and Temporal Representations for Multipitch Estimation of Polyphonic Music

News & Changelog

To view the full changelog, please go to CHANGELOG.md

version 0.3.1 (24 Dec 2021):

  1. Added VQT feature #113

version 0.3.0 (19 Nov 2021):

  1. Changed module naming. nnAudio.Spectrogram will be replaced by nnAudio.features in the future releases. Currently, various spectrogram types are accessible via both methods.

How to cite nnAudio

The paper for nnAudio is avaliable on IEEE Access

K. W. Cheuk, H. Anderson, K. Agres and D. Herremans, "nnAudio: An on-the-Fly GPU Audio to Spectrogram Conversion Toolbox Using 1D Convolutional Neural Networks," in IEEE Access, vol. 8, pp. 161981-162003, 2020, doi: 10.1109/ACCESS.2020.3019084.

BibTex

@ARTICLE{9174990, author={K. W. {Cheuk} and H. {Anderson} and K. {Agres} and D. {Herremans}}, journal={IEEE Access}, title={nnAudio: An on-the-Fly GPU Audio to Spectrogram Conversion Toolbox Using 1D Convolutional Neural Networks}, year={2020}, volume={8}, number={}, pages={161981-162003}, doi={10.1109/ACCESS.2020.3019084}}

Call for Contributions

nnAudio is a fast-growing package. With the increasing number of feature requests, we welcome anyone who is familiar with digital signal processing and neural network to contribute to nnAudio. The current list of pending features includes:

  1. Invertible Constant Q Transform (CQT)
  2. CQT with filter scale factor (see issue #54)
  3. Speed and Performance improvements for Griffin-Lim (see issue #41)
  4. Data Augmentation (see issue #49)

(Quick tips for unit test: cd inside Installation folder, then type pytest. You need at least 1931 MiB GPU memory to pass all the unit tests)

Alternatively, you may also contribute by:

  1. Refactoring the code structure (Now all functions are within the same file, but with the increasing number of features, I think we need to break it down into smaller modules)
  2. Making a better demonstration code or tutorial

Dependencies

Numpy >= 1.14.5

Scipy >= 1.2.0

PyTorch >= 1.6.0 (Griffin-Lim only available after 1.6.0)

Python >= 3.6

librosa = 0.7.0 (Theoretically nnAudio depends on librosa. But we only need to use a single function mel from librosa.filters. To save users troubles from installing librosa for this single function, I just copy the chunk of functions corresponding to mel in my code so that nnAudio runs without the need to install librosa)

Other similar libraries

Kapre

torch-stft