A Software Framework for Musical Data Augmentation

By Brian McFee, Eric J. Humphrey, Juan Pablo Bello. 2015

144 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

Robust sound event detection in bioacoustic sensor networks (2019, methodology): “Applying randomized digital audio effects to every sample in a dataset at training time often reduces overfitting without any extra computational cost at prediction time [95].”
Deep Learning for Audio Signal Processing (2019, background): “Pitch shifting has also been shown useful for chord recognition [80], and combined with time stretching and spectral filtering for singing voice detection [81] and instrument recognition [82].”
Open-Source Practices for Music Signal Processing Research: Recommendations for Transparent, Sustainable, and Reproducible Audio Research (2019, methodology): “Data augmentation methods apply systematic perturbations to an annotated data set, such as pitch shifting or time stretching, to induce these properties as invariants in the model [16].”
Crepe: A Convolutional Representation for Pitch Estimation (2018, background): “The robustness of the model could also be improved by applying pitch-shifts as data augmentation [29] to cover a wider pitch range for every instrument.”
Evaluation of CNN-based Automatic Music Tagging Models (2020, methodology): “All employed deformations are based on an existing music data augmentation framework (MUDA) 8 [10]: Dynamic range compression following speech and music (standard) settings of Dolby E standards [23].”
CNN-Based Learnable Gammatone Filterbank and Equal-Loudness Normalization for Environmental Sound Classification (2020, methodology): “To avoid over-ﬁtting, where the network does well in predicting the training data but not the test data, audio data augmentation [14], [33], dropout [34], weight decay, and mix-up [35] were conducted.”
Scaper: A library for soundscape synthesis and augmentation (2017, background): “the transformation/manipulation of existing training data in order to generate new labeled samples [7, 8, 9].”
Chirping up the Right Tree: Incorporating Biological Taxonomies into Deep Bioacoustic Classifiers (2020, methodology): “We apply random digital audio effects as data augmentation: pitch shifting, time stretching, and addition of background noise [35].”
Invariances and Data Augmentation for Supervised Music Transcription (2017, methodology): “Our dataset augmentation techniques are inspired by analogous transformations introduced by the vision community for image classification [17, 18] and extensions of these ideas to audio [19].”
End-to-End Environmental Sound Classification using a 1D Convolutional Neural Network (2019, methodology): “Salamon & Bello (2017) presented a method based on a 2D CNN with ﬁve layers (SB-CNN) where new training samples are generated using data augmentation methods such as time stretching, pitch shifting, dynamic range compression or adding background noise (McFee et al., 2015).”