Huge would be a way to index all the subtitles of YouTube and make them indexable. This would be a massive undertaking. The video’s that do not have subtitles would need speech recognition software to get one.
A middle solution would be something I’m currently aiming at. It consist of the subtitles of a limited amount of communities that are a combination of data science and curating by hand. Only the video’s that contain subtitles would be indexed.
Small would be to focus on one particular community. This would probably be a community that has already been mapped by external parties (such as the alt-right).