Deep learning approaches allow us to look at signal processing problems from a different angle, which is currently widely disregarded in the music business. Audio is intrinsically more time-sensitive than film. You can never assume that a pixel in a spectrogram belongs to a single object. Due to the fact that audio is always transparent, spectrograms show all audible sounds overlapping in the same frame. It has been demonstrated that spectrograms can be processed as images and neural style transfer can be performed with CNNs, although the results have not been as exact as they have been for visual images. We should focus our efforts on developing more accurate representations.

Keywords : Autoencoders, Autocorrelogram, Encoding, Audio Encoders, RNN Autoencoder, Audio Frequency, Auto Correlation And Convolution, Cross Fold Validation