Augmentation Methods on Monophonic Audio for Instrument Classification in Polyphonic Music

Kratimenos, Agelos; Avramidis, Kleanthis; Garoufis, Christos; Zlatintsi, Athanasia; Maragos, Petros

doi:10.23919/eusipco47968.2020.9287745

Cited by 15 publications

(21 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…On the other hand, as we see in Table 3, 1D convolutional models are capable of extracting the most discriminative features from raw waveforms, almost as well as 2D convolutional models that work on spectrogram inputs [16,4]. Furthermore, removing the dense layers not only reduces the number of model trainable parameters, and thus the training time, but also increases the accuracy substantially.…”

Section: Architecture Comparisonmentioning

confidence: 94%

“…While traditional research, partially due to the challenge of labeling multi-instrumental music, focused upon monophonic audio [14], recent studies address polyphonic tasks, relying on the efficiency of deep learning models. Specific points of focus include investigation of the optimal input temporal resolution [15,16] and the design of the convolutional filters involved [17], while we have also experimented with sophisticated augmentation methods, attempting to isolate timbre-like characteristics [4].…”

Section: Related Workmentioning

confidence: 99%

“…However, they can provide useful results when applied to 1D signals as well [20]. We base our CNN on the architecture used in [4] that yielded strong results on the IRMAS Dataset. In order to adjust to the different input form, we substitute 2D with 1D Convolutional layers and fine-tune the pooling parameters to handle the high input dimensionality.…”

Section: Convolutional Networkmentioning

confidence: 99%

“…4. Instrument-wise performance of the proposed model and the monophonic [4] in terms of F1-score. To optimally combine the temporal information extracted from the RNN and the already utilized spatial characteristics drawn from the residual network, we additionally assess the performance of the combined networks described in Sec.…”

Section: Architecture Comparisonmentioning

confidence: 99%

“…handling the high input dimensionality of the waveforms, mining temporal features and preserving their low-level spatial locality, while reducing the computational cost of the process. We propose a lightweight end-to-end classifier for Instrument Classification that shows comparable performance to state-of-the-art spectrogrambased architectures, including our previous work on the task [4].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Deep Convolutional and Recurrent Networks for Polyphonic Instrument Classification from Monophonic Raw Audio Waveforms

Avramidis

Kratimenos

Garoufis

et al. 2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Self Cite

View full text Add to dashboard Cite

Sound Event Detection and Audio Classification tasks are traditionally addressed through time-frequency representations of audio signals such as spectrograms. However, the emergence of deep neural networks as efficient feature extractors has enabled the direct use of audio signals for classification purposes. In this paper, we attempt to recognize musical instruments in polyphonic audio by only feeding their raw waveforms into deep learning models. Various recurrent and convolutional architectures incorporating residual connections are examined and parameterized in order to build end-to-end classifiers with low computational cost and only minimal preprocessing. We obtain competitive classification scores and useful instrumentwise insight through the IRMAS test set, utilizing a parallel CNN-BiGRU model with multiple residual connections, while maintaining a significantly reduced number of trainable parameters.

show abstract

Section: Architecture Comparisonmentioning

confidence: 94%