Convolutional neural networks for the classification of guitar effects and extraction of the parameter settings of single and multi-guitar effects from instrument mixes

Abstract Guitar effects are commonly used in popular music to shape the guitar sound fit specific genres, or create more variety within musical compositions. The not only is determined by choice of effect, but also heavily depends on parameter settings effect. Previous research focused classification and extraction their from solo audio recordings. However, realistic instrument mixes. This work investigates use convolution neural networks (CNNs) for samples containing guitar, bass, keyboard, drums. CNN was compared baseline methods previously proposed, like support vector machines shallow together with predesigned features. On two datasets, achieved accuracies $$1-5\,\%$$ 1 - 5 % above accuracy, achieving up $$97.4\, \%$$ 97.4 accuracy. With values between 0.0 1.0, mean absolute errors below 0.016 distortion, 0.052 tremolo, 0.038 slapback delay effect were achieved, matching surpassing presumed human expert error 0.05. approach found generalize further effects, 0.05 chorus, phaser, reverb, overdrive For sequentially applied combinations delay, slightly increased performance single range 0.1. be moderately robust noise pitch changes background instrumentation suggesting that extracted meaningful