End-to-end learning for music audio

By Sander Dieleman, Benjamin Schrauwen. 2014

419 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Speech & Audio, 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 Speech & Audio

In Signal Processing

Citations in the corpus

Unsupervised Speech Representation Learning Using WaveNet Autoencoders (2019, background): “to learn auditory-like frequency decompositions on music [23] and speech [24]–[27] in their lower layers.”
A Wavenet for Speech Denoising (2017, methodology): “rning multi-scale structure directly from raw audio, may have great potential in learning such structures. To this end, discriminative models have been used in a end-to-end learning fashion for music [6, 15] or speech classiﬁcation [5, 20, 39]. Raw audio waveforms have also been successfully employe”
GANSynth: Adversarial Neural Audio Synthesis (2019, background): “Convolu-tional ﬁlters trained for different tasks on this data commonly learn to form logarithmically-scaled frequency selective ﬁlter banks spanning the range of human hearing (Dieleman & Schrauwen, 2014; Zhu et al., 2016).”
End-To-End Source Separation With Adaptive Front-Ends (2017, background): “al [11], and Dieleman and Schrauwen [12] have”
A Simple Fusion of Deep and Shallow Learning for Acoustic Scene Classification (2018, background): “, speech recognition [8], automatic music tagging [9] or environmental sound classification [10].”
Automatic Tagging Using Deep Convolutional Neural Networks (2016, methodology): “This result is aligned with the preferences of mel-spectrograms over STFT on automatic tagging [5,6,18,27].”
Wavelet Networks: Scale Equivariant Learning From Raw Waveforms (2020, neutral): “[18] Sander Dieleman and Benjamin Schrauwen.”
Automatic Sleep Stage Scoring with Single-Channel EEG Using Convolutional Neural Networks (2016, methodology): “While this has an eﬀect similar to the stacking layer, we think that explicitly stacking the outputs of layer P2 makes clear the correspondence between CNN methods and hand-engineered feature methodologies.”
AST: Audio Spectrogram Transformer (2021, background): “E[1] E[2] E[3] E[4] E[5] E[6] E[7] E[8] P[1] P[2] P[3] P[4] P[5] P[6] P[7] P[8] Linear Output”
Freesound Datasets: A Platform for the Creation of Open Audio Datasets (2017, methodology): “4 To alleviate these issues, researchers typically use top-50 tags [25, 6], or a cleaner and pre-processed version (Magtag5k [21]).”
Acoustic Scene Classification by Ensembling Gradient Boosting Machine and Convolutional Neural Networks (2017, background): “, speech recognition [9], automatic music tagging [10] or environmental sound classification [11].”
End-to-End Environmental Sound Classification using a 1D Convolutional Neural Network (2019, background): “…(Sajjad Abdoli), patrick.cardinal@etsmtl.ca (Patrick Cardinal), alessandro.koerich@etsmtl.ca (Alessandro Lameiras Koerich) 1 Corresponding author & Schrauwen, 2014), large-scale video clip classiﬁcation based on audio information (Hershey et al., 2017), music genre classiﬁcation (Costa et al.,…”
Multi-Level and Multi-Scale Feature Aggregation Using Pretrained Convolutional Neural Networks for Music Auto-Tagging (2017, methodology): “We filtered out the tags and used most frequently labeled 50 tags for both MTAT and MSD, following the previous work [2], [4].”
Deep Learning for Audio Signal Processing (2019, background): “pply to a full recording, without timing information. Bridging the gap from an input sequence to global labels has been approached in different ways, which are instructive to compare. Dieleman et al. [109] train a CNN with short 1D convolutions (i.e., convolving over time only) on 3-second log-mel s”
Sample-level Deep Convolutional Neural Networks for Music Auto-tagging Using Raw Waveforms (2017, methodology): “Since the time-frequency representation is two dimensional data, previous work regarded it as either two-dimensional images or one-dimensional sequence of vectors [11,13–15].”
Multichannel Signal Processing With Deep Neural Networks for Automatic Speech Recognition (2017, background): “The mic array is placed at the center of the room and closer to the wall, respectively, to capture reasonably different reverberation characteristics.”
Learning Multiscale Features Directly from Waveforms (2016, background): “A popular approach is pass the waveform through strided convolutions, or networks connected to local temporal frames, often followed by a pooling step to create invariance to phase shifts and further downsample the signal [3, 4, 5, 6, 7, 8].”
Speaker identification and clustering using convolutional neural networks (2016, methodology): “We then perform dynamic range compression of the spectrograms by applying the elementwise function f(x) = log(1 + C · x) as in [21], with C = 10′000.”
Invariances and Data Augmentation for Supervised Music Transcription (2017, background): “Recent work shows that end-to-end architectures construct a first-layer feature representation that is qualitatively comparable to classical frequency filterbank transforms such as the STFT or CQT [8, 7].”