search

KinWaiCheuk / nnAudio

Audio processing by using pytorch 1D convolution network
MIT License
1.03k stars 89 forks source link
1d-convolution audio-processing cqt-spectrogram melspectrogram neural-network preprocessing pytorch spectrogram spectrogram-conversion-toolbox stft

readme

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)

(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. 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