Minimum number of training steps for unconditional synthesis

[bharani-y]

Hi,

This is a great library, I am training SoundStorm on the Librispeech 1000hrs dataset and want to know how many training steps are required to start hearing some sensible audio using generate function, currently, it is trained for 100K steps and the audio is still pure noise, can you specify after how many steps you started to hear some sensible audio?

Thanks