I'm working on adding a feature so that every search result can be subscribed to as a podcast as well, so you can listen to everything on one topic by different people.
I'd love that podcast subscription capability, though listening to it in HTML5 means I can speed up the playback without having to download the audio file first (my podcast app, PocketCast, doesn't let me do this).
A while ago, I was thinking about doing a search engine on top of audio. The startup deepgram seemed to offer interesting tools, but didn't get a chance to play with it: https://www.deepgram.com/
Currently I have it searching transcripts, but I want to get machine transcriptions in as well, because it should work better how often someone says 'um', which I'm going to try as one of the ranking factors.