What I'm really looking forward to is better free
speech recognition libraries. (Current software produces output that's good for finding certain keywords, but output text is just an unreadable stream of words with like 20% of the words wrong or missing, and it's especially bad when you have dialog between multiple different, often overlapping voices...)
I could then write a script that not only produces an accurate
transcript of the show, but can also hyperlink each line to the podcast time, automatically split the show by segment / caller, etc. That would result in a completely different listening experience - jump to what interests you the most, skip around, skim through some parts, etc.
This can be combined with tagging, ratings, and wikification to add additional meta-data into the transcript. You could then just query a script to produce an RSS feed of all segments with a particular callers, topics, or keywords for the past X weeks!