Build An Audio AI App Transcripts
Chapter: Feature 3: Summarize
Lecture: Introduction to LLM Summaries

Login or purchase this course to watch this video and the rest of the course contents.
0:00 We're taking it to the next level with our audio work. We're not just asking for transcripts and feeding those into search, which already is awesome.
0:11 We're going to go and use the Lemur LLM that understands audio to work with this transcript
0:18 and this audio file the way you would think of maybe something with ChatGPT.
0:23 But in fact, it's quite a bit better because it understands very large documents and it
0:28 can keep a lot of tokens and have a large context for all the information it works with. I think you're going to be pretty impressed.
0:36 So we're going to use our LLM in this chapter for two things, kind of the same two sides of the same thing.
0:44 We're going to go to an episode like this one that we've been working with throughout our examples.
0:49 And once we have the transcript, we're going to be able to say, why don't you just go ahead and summarize that for me. We want the summary in two ways.
0:57 We want this TLDR. This is really valuable because as a consumer of this podcast, you know, you can see it's 55 minutes and six seconds.
1:08 That's a lot of time. Even if I listen at a double speed, that's still a lot of time and mental energy and focus.
1:14 Here's one paragraph that I can read and decide. Is that worth going into? Is that worth spending the time? The other part will be the key takeaways.
1:23 You could use this before you listen or even after. You've listened to the episode.
1:27 You're like, oh, I know there was a couple of things that gave some advice. What was it? Boom, boom, boom.
1:33 You can look in there and see what those are. We're going to generate that with the Lemur LLM from Assembly AI.
1:39 And of course, we could take those things, fold them back into how we might use the transcript.
1:45 We could go into the transcript and search for those things and try to get the whole context.
1:49 And we'll take both the TLDR and the key moments and feed those into our search, making our search even better and even deeper than it was before.
1:59 It's going to be really fun.

Talk Python's Mastodon Michael Kennedy's Mastodon