Create hq_dataset

[rebotnix]

This is a first grab sample for an HQ/HD Dataset where we collect several youtube videos and cut out the speech part.

[prajwalkr]

Hello, thank you very much for your interest and contribution. We do not want to add any code related to the high-resolution model, until there is a trained, working high-resolution Wav2Lip. The reason is that we are anticipating many other technical challenges along the way (some of them could be might need extensive experimentation/research themselves), and until then, we feel it is best to maintain this repo to be the official implementation of the published paper.