These might be useful augmentations in the audio domain: (see "An Ensemble of Convolutional Neural Networks for Audio Classification" https://www.mdpi.com/2076-3417/11/13/5796/htm section 4.2)
[ ] clipping: normalize such that 10% of samples outside [-1,1] then clamp
[ ] harmonic distortion applyHarmonicDistortion is the repeated application of quadratic distortion to the signal; in this study, the following distortion was applied five consecutive times: sout=sin (2πsin);
These might be useful augmentations in the audio domain: (see "An Ensemble of Convolutional Neural Networks for Audio Classification" https://www.mdpi.com/2076-3417/11/13/5796/htm section 4.2)
applyHarmonicDistortion is the repeated application of quadratic distortion to the signal; in this study, the following distortion was applied five consecutive times: sout=sin (2πsin);