Play It Again, Claude

I'm taking Korean this semester. The textbook came with audio — every conversation and every narration in the book, recorded by native speakers. Exactly what you want when you're trying to learn to hear a language.

I downloaded all the audio files in a folder. Lots and lots of files; it seems like there were a couple hundred of them, named in some scheme that meant something to whoever built the download, but nothing to me. To hear the conversation from Lesson 3, I had to sit at my desk and click through files until I found it. Some files had four sentences run together in one recording. I had the exact material I needed and no practical way to study with it.

So I described the problem to Claude Code the way I'd describe it to a person. Here's a pile of audio, here's the textbook it belongs to, help me make this usable.

It went after the run-together recordings first, and that was its idea, not mine. It used a tool called FFmpeg to find the pauses inside each recording and cut there, so every sentence became its own clip. Then it went through every clip it had made: listened to it, wrote down what was being said in Korean, and filed it by lesson and section. Then it built a web page. Every line of the conversation in Korean, with a play button beside it. A button to play one speaker's whole turn. A button to play the entire conversation. Speakers color-coded so you can follow who's talking. It's hosted on GitHub for free, so on my phone, all I need is an Internet connection and I'm off to the races.

It took about an hour to do all of this; start to finish.

The part I didn't expect came later. I stopped listening to music at the gym and started running Korean conversations instead. Which was great, except I had to keep pulling my phone out to press the button again and again, and the audio stalled between lines because it wasn't loading and storing the audio files ahead of time.

So I went back to Claude and told him exactly what was happening - what was annoying me. It added a loop with a three-second breath between passes, made the page download all the audio files in advance, and added a stop button, since until that moment there was no way to stop the thing once it started speaking to me!

Here's the part I glossed over. I could have labeled the first few clips myself. 안녕하세요 I know. But the audio runs past my vocabulary somewhere in Lesson 2, and by the back of the book I can't tell you where one word ends and the next one starts, never mind spell it. This wasn't a job I could have done slowly. It was a job I could not have done.

So it didn't just cut and sort and name. It listened to a language I'm still learning to hear and wrote down what was said, in Korean, and put that text on the screen next to the play button. That turned out to be the whole thing. I press play, I read along, and the sounds start sticking to my brain. I asked for an organized folder. What came back was a tutor that had to build itself first, and it's waiting on my phone every time I get to the gym.