The problem I currently have is that I want to train an AI to produce music, like music that contains voices etc... However, the problem is that with a WAV file, one second of audio can be up to 48,000 inputs, which is extremely detrimental to the ai's learning process and prevents it from really gaining any knowledge about context. I've tried to do the fast Fourier transformation, but the amount of data coming in varies depending on what part of the song I'm training it on, which will not work since I can't know ahead of time what every single time unit of every single song will have! And the max number of inputs is still 24,000
Is there any other way of compressing data down in a way that I can give my ai something within the range of a couple hundred inputs for a second or two?