Improving music source separation based on deep neural networks through data augmentation and network blending

By Stefan Uhlich, M. Porcu, Franck Giron, Michael Enenkl, Thomas Kemp, Naoya Takahashi, Yuki Mitsufuji. 2017

226 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Music Source Separation in the Waveform Domain (2019, methodology): “We also selected MMDenseLSTM [Takahashi et al., 2018], a multi-band dense net with LSTMs at different scales of the encoder and decoder.”
Learning to Compose Domain-Specific Transformations for Data Augmentation (2017, background): “In addition, various studies have applied heuristic data augmentation techniques to modalities such as audio [31] and text [20].”
Wave-U-Net: A Multi-Scale Neural Network for End-to-End Audio Source Separation (2018, methodology): “To alleviate the problem of ﬁxed spectral representations widely used in previous work [6, 11, 13, 14, 20, 23], an adaptive front-end for spectrogram computation was developed [24] that is trained jointly with the separation network, which operates on the resulting magnitude spectrogram.”
End-To-End Source Separation With Adaptive Front-Ends (2017, methodology): “Several neural network (NN) architectures and methods have been proposed for supervised single-channel source separation [1, 2, 3, 4].”
Further Advantages of Data Augmentation on Convolutional Neural Networks (2018, methodology): “In domains other than computer vision, data augmentation has also been proven effective, for example in speech recognition Jaitly & Hinton (2013), music source separation Uhlich et al. (2017) or text categorization Lu et al. (2006).”
Denoising Auto-Encoder with Recurrent Skip Connections and Residual Regression for Music Source Separation (2018, methodology): “[15] showed that data augmentation is crucial to compensate for the scarcity of training data in music source separation.”
Improving singing voice separation using Deep U-Net and Wave-U-Net with data augmentation (2019, background): “[13] proposes rather basic strategies for data augmentation for”
D3Net: Densely connected multidilated DenseNet for music source separation (2020, methodology): “The network outputs are used to calculate the multichannel Wiener ﬁlter (MWF) to obtain the ﬁnal separations, as commonly performed in frequency domain audio source separation methods [3,8,10,11].”
Data augmentation instead of explicit regularization (2018, background): “Although it is most popular in computer vision, data augmentation has also proven effective in speech recognition (Jaitly & Hinton, 2013), music source separation (Uhlich et al., 2017) or text categorization (Lu et al.”
The 2018 Signal Separation Evaluation Campaign (2018, unknown): “They are: UHL3 [28], TAK2 [26], TAK3 [27], TAU [27, 28].”
Singing Voice Separation: A Study on Training Data (2019, methodology): “When training on a small dataset like MUSDB, data augmentation is regularly cited as a way to improve separation performances [8].”
A Kernel Theory of Modern Data Augmentation (2018, methodology): “As a testament to its growing importance, the technique has been used to achieve nearly all state-of-the-art results in image recognition [1, 6, 9, 23], and is becoming a staple in other areas as well [30, 15].”
Upsampling Artifacts in Neural Audio Synthesis (2020, methodology): “Like the original Demucs, we use: very large models, of ≈700M parameters [3]; weight rescaling, so that input and output signals are of the same order of magnitude [3]; and their data augmentation schema, creating new mixes on-the-fly [3, 27].”
Open-Unmix - A Reference Implementation for Music Source Separation (2019, methodology): “…user to use the model right away or fine-tune it on user-provided data (Stöter & Liutkus, 2019b, 2019c) Open-Unmix is based on the bi-directional LSTM model from (Uhlich et al., 2017) and we compared it to other separation models that were submitted to the last SiSEC contest (Stöter et al., 2018).”
Mmdenselstm: An Efficient Combination of Convolutional and Recurrent Neural Networks for Audio Source Separation (2018, methodology): “Short-time Fourier transform magnitude frames of the mixture, windowed at 4096 samples with 75% overlap, with data augmentation [12] were used as inputs.”