road2018 / torchsubband

Pytorch implementation of subband decomposition

torchsubband

This's a package for subband decomposition.

It can transform waveform into three kinds of revertable subband feature representations, which are potentially useful features for music source separation or similar tasks.

Usage

Installation

pip install torchsubband

A simple example:

from torchsubband import SubbandDSP
import torch

# nn.Module
model = SubbandDSP(subband=2) # You can choose 1,2,4, or 8 
batchsize=3 # any int number
channel=1 # any int number
length = 44100*2 # any int number
input = torch.randn((batchsize,channel,length))

# Get subband waveform
subwav = model.wav_to_sub(input)
reconstruct_1 = model.sub_to_wav(subwav,length=length)

# Get subband magnitude spectrogram
sub_spec,cos,sin = model.wav_to_mag_phase_sub_spec(input)
reconstruct_2 = model.mag_phase_sub_spec_to_wav(sub_spec,cos,sin,length=length)

# Get subband complex spectrogram
sub_complex_spec = model.wav_to_complex_sub_spec(input)
reconstruct_3 = model.complex_sub_spec_to_wav(sub_complex_spec,length=length)

Reconstruction loss

The following table shows the reconstruction quality. We tried a set of audio to conduct subband decomposition and reconstruction.

Subbands	L1loss	PESQ	SiSDR
2	1e-6	4.64	61.8
4	1e-6	4.64	58.9
8	5e-5	4.64	58.2

You can also test this program by running the following test script. It will give you some evaluation output.

from torchsubband import test
test()

Citation

If you find our code useful for your research, please consider citing:

    @misc{liu2021cwspresunet,
        title={CWS-PResUNet: Music Source Separation with Channel-wise Subband Phase-aware ResUNet},
        author={Haohe Liu and Qiuqiang Kong and Jiafeng Liu},
        year={2021},
        eprint={2112.04685},
        archivePrefix={arXiv},
        primaryClass={cs.SD}
    }

    @inproceedings{Liu2020,   
      author={Haohe Liu and Lei Xie and Jian Wu and Geng Yang},   
      title={{Channel-Wise Subband Input for Better Voice and Accompaniment Separation on High Resolution Music}},   
      year=2020,   
      booktitle={Proc. Interspeech 2020},   
      pages={1241--1245},   
      doi={10.21437/Interspeech.2020-2555},   
      url={http://dx.doi.org/10.21437/Interspeech.2020-2555}   
    }