Cutting audio by crossing out words

I’ve set something up on my computer at home that I’m rather chuffed about.

As you may have seen, I’ve recently been recording conversations as an exchange of audio recordings – I send someone a question, they send me a reply, and so on. (Like these, with Andrew Clover and Michael G.)

Well recently I wanted to stitch together the recordings into a single conversation. I was away from my desktop computer and wanted to do it on my phone – also make the combined recording shorter.

I’ve now set up a system in which the computer listens to the recording and transcribes every word – for me to read it through, then cross out the bits I don’t want. A minute or two later it gives me a new recording with those bits gone.

What’s clever about this is that I can do the crossing out on my phone, wherever I happen to be, and the computer at home does the rest.

This is yet another example of creativity coming out of constraint. I built this because I was away from home last week.

I had a conversation I wanted to shorten and only my phone to do it on. The conversation was with Ben Akers, and you can hear it here: How we first met. It was four voice notes on WhatsApp, two questions from me and two answers from him, and it ran to 7 minutes 21 seconds. With help from an AI, Claude, I botched together a way of cutting it by crossing out words, and it came out at 5 minutes 14 seconds.

I didn’t want to be dependent on Claude to do it all again, so I instructed Claude to build a system. And that’s now in place.

Mind you, I haven’t tried the new version yet. So I can’t promise it works. I’ll let you know.

What follows is a note to myself, so that I remember how to use it. I’m keeping it here, in public, because somebody else might like to try something similar – most likely somebody like me, who makes things, enjoys playing around with tech, and is interested in conversation.

***

To make a recording shorter

  1. Find the recording on the phone.
  2. Tap Share.
  3. Tap Level & Transcribe.
  4. Do something else for a few minutes. Enjoy myself.
  5. Open Subtext (a free App on my phone). Go to iCloud Drive, then Desktop, then Podcast inbox.
  6. Open the file with “cut sheet” in its name.
  7. Ignore the top few lines. Below the row of equals signs is the transcript: every word that was said.
  8. Delete the words I don’t want.
  9. Save.
  10. Do something else for two minutes. Eat a biscuit!
  11. Open the Files app. Tap iCloud Drive, then Desktop, then Podcast inbox. Play the file whose name ends in “cut”.

If it isn’t right, go back to step 6 and change the cut sheet again.

If I want to be able to see what I’ve cut, don’t delete the words. Put two squiggles either side of them instead, ~~like this~~.

If a word sounds chopped, go back to the cut sheet and delete one word more, or one word less, next to it.

To join several recordings into one

  1. Open the Files app. Tap iCloud Drive, then Desktop, then Podcast inbox.
  2. Make a new folder there. Call it whatever the finished piece is called.
  3. Move the recordings into the folder.
  4. Do something else for a few minutes.
  5. Open Subtext. Go to the new folder, and open the file with “join sheet” in its name.
  6. Below the row of equals signs there is one line for each recording.
  7. Move the lines into the order I want to hear them.
  8. Save.
  9. Do something else for two minutes.
  10. Go back to the Files app. Play the file whose name ends in “joined”.

To make one of the recordings shorter, it has its own cut sheet in the same folder. Start at step 6 of the first list.

If nothing happens

  1. Is the Mac at home switched on? It needs to be, because it’s the Mac that does all the work, not Claude.
  2. Still nothing? Ask Claude. It built the thing and kept notes.

The programs underneath are free, made by people who share them: Whisper and Parakeet do the listening, and ffmpeg does the cutting.

-

If you enjoyed this post, please consider sharing it by email or on social media. Here's the link. Thanks, JPF.