The following is a GAN that generates Frequency domain Audio signals. Currently it generates 1 second of 8KHz audio, represented as a 2X4000 matrix of frequencies strengths real and imaginary signals.
The generated frequencies, when converted to time-domain, sound like noise at first, but that will change to tunes and/or beeps. but not yet to music or words. I have also tried time domain, also to no avail.
To install, install Tensorflow, imageio, ffmpeg, and run imageio.plugins.ffmpeg.download()
We also provide two helper scripts: install-dependencies.sh for installing the necessary dependencies.