Creating AI-Driven Transcript Website for Podcasts (Full Transcript)

Speaker 1: How's it going everybody? Today I'm gonna show you what is possible to do with artificial intelligence or more specifically OpenAI's Visper models. So what I did was I created a website in React and this is for the Jocko podcast which is hosted by Jocko Willink, a former Navy SEAL. If you don't know anything about him you can check out his books or his social media. But in this case I created for each episode I have the title of the podcast episode, I have the date it was published, I have some tags and I have a description. This all comes from the YouTube video. So in this case this one. And then what I did is I summarized each episode with the Visper models. And this is based on the text from Visper models but this is just normal natural language processing. Then I have a word cloud of the most common words used in the podcast episode. And then I have the transcript. And then I have the transcript of each episode here. So this work is actually inspired by Andrej Karpathy and Aleksa Gordik. So what Andrej did was he created a transcript website for the Lex Friedman podcast. And he also used the Visper captions or the Visper model. In his case he has timestamps. Which is clickable. You can click it. And then you get the video with the correct timestamp when the ad is finished. And he actually talked about this in a podcast episode with Lex Friedman. So if you want to actually check out the clip where he talks about OpenAI as Visper. I recommend the clip. But then it's my website is also inspired by him. This is also inspired by Aleksa. So what he did is he has Andrew Huberman transcripts website. And in his case he has here AI generated summaries of breakdowns of what is happening in the podcast. He previously had the transcripts here also. But it's no longer here. It's maybe. Yeah. Okay. If you change it to raw transcripts you get the timestamps with what was said. Then if you check out chapter summaries and then episode summaries. So yeah. This is what he did. And then here's what I did. In my case I'm going to show you. What I did. So the entire thing is hosted on GitHub. If you want to check it out yourself. So here's the repository for the transcripts part where I take download the audio. Then take each episode of the audio and I transcribe it with the whisper models. But then I host the entire thing. In this repository called choco podcast website. So it's a separate repository for what the purpose is. This is just to transcribe the audio. And this is just to host the website. And I'm going to show you the code. Console. So here's how it works. It's. You run this part here. To install whisper. You run this part to download each. Audio file. So in this case. This is actually. The. This playlist is the playlist. And what it does is. It downloads. The entire playlist. To the folder that the file is in. So if you're going to copy what I did or use my code. What I actually suggest doing is. Especially if you're on Linux. I would have this file here. Inside of the file folder you want the audio files to be in. So this would download the entire audio files. But then. Because down below. I'm going to write the. Or take each audio file. In Linux. If it's in a sub directory. It will fail. But if it's. In Windows. It doesn't matter. Where the audio files is. Which is. Weird. For some reason why that's happening. But. Yeah. Just run the entire file. Inside of the folder. You want it to be in. And. This code here. What this code does. This part. Is. It takes. The. Text. And it creates. A timestamp for it. So. Like. Here you can see. At. Zero. Seconds. To. Two seconds. This is what was said. And then. It goes down. And then. It goes up. And then. It goes down. And then. It goes up. And then. And then. But. This is. This part here. Like. Creating this. It's not actually. My. My code. I. Got this code from. This. Here. Tutorial. So. It's just. Getting started. With. OpenIS Whisper. And. Using. That code. You can. Just. Get the file. Yourself. It makes it. Much. Easier. To start. But then. What I did. Is. I ran. Most of the episodes. With. A tiny model. But. If you have a good. GPU. You can run it. With. The base. Or large. What that means. Is. If. We. Look. At. This. So. We. There's. A tiny model. The tiny model. Has. 39. Million. Parameters. There's. A small model. Which. Has. 244. Million. Parameters. The. Medium. Has. 769. Million. Parameters. And. The. Large. Has. 1550. Million. Parameters. Or. It. Means. That. It. Was. Trained. On. This. Or. The. 40. Series. With. More. Than. 10. Gigabit. Ram. With. Your. Ram. You. Can. Run. This. Easily. Or. Not. Easily. But. You. Can. Run. This. If. You. Have. On. A. 3080 TI. For. Me. Around. Two. Minutes. Maybe. Two. And. A. Half. Minute. But. The. Large. Model. Actually. Ran. In. I. Believe. It. Was. Three. Minutes. 50. Seconds. You. Don't. Have. That. Many. Audio. Files. And. You. Have. A. Good. GPU. Because. It. Took. I. Believe. 20. Hours. To. Run. 370. Episodes. Or. Like. 340. Episodes. On. Tiny. The. Large. Model. But. What's. Impressive. With. The. Whisper. Model. Is. It. Actually. Supports. Multiple. Languages. And. It. Just. Picks. It. Up. Automatically. What's. Happening. And. In. This. Case. Understanding. What's. Happening. With. Icelandic. Language. Which. Is. Kind. Of. Unbelievable. How. That. Actually. Works. But. If. You. Want. To. Read. More. About. What. The.

Speaker 2: Whisper. Models. Are. You. Like.

Speaker 1: Or. Maybe. Like. The. Accent.

Speaker 2: Scottish. Accent. So. It's. Actually. Amazing. How. It.

Speaker 1: Understands. Noise. And. Can. Reduce. The. Sentences. From. That. Which. Is. Pretty. Impressive. So. Yeah. I. Was. Using. The. Multilanguage. Multitask. Supervised. Data. Collected. From. The. Web. But. I'm. Going. To. Run. You. Through. The. Code. What. I. Did. So. Using. The. Tiny. Or. What. That. Format. Is. But. It. Actually. This. Was. Just. Done. To. This. Part. Is. Printing. Out. The. Title. Of. The. File. Name. Without. The. File. Ending. But. Then. Your. Text. Is. Isn't. Separated. But. Then. I. Have. The. Caption. Files. Which. Is. Timestamped. That's. All. Happening. Here. So. Yeah. When. You're. Done. With. These. Few. Lines. Of. Codes. Of. Captions. The. Audio. File. Captions. There. But. Then. What. I. Did. Was. Here's. Just. Some. Help. Functions. This. Was. Mostly. Done. Because. The. File. Names. Of. The. Files. It. Didn't. Understand. These. Episodes. Well. So. I. Had. To. Manually. Fix. That. Here. But. What. Happening. Here. Is. I. Take. Out. The. Weird. Symbols. And. Actually. Just. Create. Clean. Files. Here. Clean. So. Here. I've. Changed. The. Entire. Files. To. Better. Represent. What's. Actually. Happening. I. Can. Just. Rename. Copy. Just. To. Show. You. The. Difference. So. Here's. The. And. Lower. Case. So. That's. Just. What's. Happening. Here. And. This. Helper. Functions. Then. Move. I. Copy. The. File. With. The. Correct. File. Names. Into. Our. Clean. Directory. Here. And. Then. What. I'm. Doing. Here. Is. For. The. Website. I'm. Creating. Links. So. Like. You. See. Here. This. Is. If. I. Do. Inspect. Element. On. The. Here. And. For. The. One. That's. Just. What's. Happening. Here. I. Do. The. Absolute. And. The. Absolute. Number. And. Then. This. Part. Isn't. Used. Right. Now. I. Forgot. To. Include. It. This. Just. Of. The. AI. Here. I. Put. That. Into. Folder. And. I. Create. The. Work. Cloud. Here. Too. So. The. Most. Common. Words. It's. All. Happening. Here. Can. Just. Go. Through. Yeah. I. Had. To. Use. Jason. File. To. Get. The. To. Get. This. Like. This. Image. Get. This. Title. The. When. It. Happened. And. The. Tags. And. The. Discussion. The. Videos. Here. So. It's. Just. A. Jason. File. I. Don't. Know. What's. Happening. Yep. Just. A. Jason. File. Yeah. What. Do. You. Have. To. Do. Probably. If. Jason. Something. And. Then. You. Get. I. Believe. I. Don't. See. It. YouTube. Made. I. Will. Include. A. Link. To. The. Website. I. Used. To. This. Because. You. Can. But. You. Can. Actually. Do. It. If. You. Want. Was. Just. Much. Easier. To. Not. That. Then. What. I. Did. Was. I. Put. This. Entire. Thing. Into. Folders. So. All. The. Captions. Of. That. Episode. Is. The. Full. Text. Of. Episode. One. Is. The. Jason. Just. Informations. About. The. Episode. Is. Links. I. Don't. Use. This. But. I. Could. Here. A. Summary. Is. A. Thumbnail. So. Just. A. Link. To. The. Thumbnail. Is. The. Titles. Of. Each. Episode. This. Is. Oh. Yeah. I. Ask. I. Skipped. This. Part. Because. I. Need. It. But. Then. Here's. The. Work. Cloud. Images. Of. Each. Episode. So. This. Is. The. Structure. Of. How. The. Episode. Directory. Looks. Like. But. Then. If. We. And. If. We. Take. A. Look. At. The. Website. Like. The. Public. Now. The. Pages. Here. I. Just. Have. The. Results. And. The. Results. Come. From. Components. Results. Somebody. On. Yeah. For. The. Mouse. Okay. Here's. The. Results. Like. You. See. This. Is. Flip. Move. Means. Just. The. This. Part. Like. If. I. Hover. That's. The. That's. This. Part. So. It's. Each. Episode. Is. Inside. Of. The. Array. Of. Videos. Here's. The. Thumbnail. Part. To. Get. The. Thumbnail. Get. The. Title. Get. Everything. Is. The. Navigation. Part. The. Header. Part. But. Then. Inside. The. Episode. I'm. Getting. Everything. From. The. Text. Files. Or. The. JSON. File. So. This. Is. Just. How. I. Built. The. Website. Around. The. Transcripts. From. The. Content. Of. The. Video. And. Subscribe. For. More. Videos. In. The. Future.