Abstract

Virtual analog modeling emulates the processing characteristics of a given physical device. This has been an active field of research and commercial innovation in which two main perspectives have been historically adopted. The first one: white-box, seeks to reproduce the exact behavior through algorithmic simulation of circuits or physical phenomena. The second one: black-box, aims to learn the approximation function from examples recorded at the input and output stages of the target device. In this second approach, deep learning has emerged as a valuable strategy for linear systems, such as filters, as well as nonlinear time-dependent ones like distortion circuits or compressors.

The spring reverb is an audio effect with a very long and rooted history in music production and performance. Based on a relatively simple design, this device is an effective tool for artificial reverberation. The electromechanical functioning of this reverb makes it a nonlinear time-invariant spatial system that is difficult to fully emulate in the digital domain with white-box modeling techniques.

This thesis aims to address the modeling of spring reverb, leveraging end-to-end neural audio effect architectures through supervised learning. Recurrent, convolutional, and hybrid models have successfully been used for similar tasks, especially compressor and distortion circuit emulations. Using two available datasets of guitar recordings, we evaluate with quantitative metrics, acoustical analysis, and signal processing measurements the efficiency and the results of four neural network architectures to model this effect. We present the results and outline different strategies for this modeling task, providing a reproducible experimental environment with code.

Audio Examples

Some dry and processed audio examples are available below. The audio files are in 48kHz 24bit wav format. All the models have been trained with the same dataset, and the same training procedure with a single NVIDIA GeForce GTX 1080 Ti GPU.

Drum loop

• Dry / raw

• Processed with the GCN model

• Processed with the TCN model

• Processed with the WaveNet model

Pluck synth sample

• Dry / raw

• Processed with the GCN model

• Processed with the TCN model

• Processed with the WaveNet model

Bibtex Citation

This is not a PhD thesis, but a Master’s thesis, Zenodo does not allow to specify this in the metadata.

@phdthesis{francesco_papaleo_2023_8380480,
  author       = {Francesco Papaleo},
  title        = {Neural Audio Effect Modelling Strategies for a 
                   Spring Reverb},
  school       = {Universitat Pompeu Fabra},
  year         = 2023,
  month        = sep,
  doi          = {10.5281/zenodo.8380480},
  url          = {https://doi.org/10.5281/zenodo.8380480}
}