6 min read
Swimming with Audiobooks

I’ve been swimming a lot recently. I find swimming to be pretty boring, so I’ve been listening to music with swim headphones to occupy my mind. A few weeks ago, my cheapo headphones were stolen from the side of the pool. To replace them, I bought some Shokz OpenSwim headphones. Wow, these are so much better.

The audio quality on the old ones was bad, barely enough to hum along to songs I already knew. The Shokz are great! They are as good as the bone-conduction headphones I use on dry land. This made me wonder: could I listen to audiobooks with these?

Getting the audiobooks

Bluetooth doesn’t work through water, so I couldn’t just stream the audio from my phone. I needed to get mp3s, like in the 2000s.

I wrote a program to download audiobooks that I checked out from the library. The program

  1. Opens a web browser,
  2. Signs into the library site,
  3. Begins playing the audiobook, skipping 5 minutes ahead every few seconds,
  4. Watches for network requests for audio files and saves them in a folder.

This gives me a folder with a handful of mp3s in the right order.

Splitting the files

The shokz have no way to “scrub” and jump back a bit if I miss something. I could miss a crucial plot point when someone asks to join my lane!

There is a control to go back to the start of the track, but a “track” of an audiobook is 1-3 hours. That’s too much to go back.

What if we split each big track into smaller ones? Say, 60 seconds each. That way, I can go back just a bit at a time. This can be done with ffmpeg:

ffmpeg -i alloy-of-law/alloy-of-law-01.mp3 \
-f segment -segment_time 60 \
-c copy \

I mostly copied this from a StackOverflow answer, but I’ll do my best to interpret it. Line by line, that says

  1. Run ffmpeg on the first track of my audiobook, at alloy-of-law/alloy-of-law-01.mp3.
  2. Segment the file, 60 seconds per track
  3. Copy the codec (don’t bother to re-encode the data)
  4. Write the output to some new files, where %03d is a placeholder for the segment number.

This produced 59 files, each 60 seconds long. Wanting to try it out, I copied the files onto my headphones. The book started right in the middle of a sentence 🤦🏻.

Ordering the songs

After some research, I came across some documentation that suggested the OpenSwim headphones determined the order of songs according to when each file was copied to the memory rather than by file name. When I copied over my 59 tiny tracks, some of the later ones must have finished before the earlier ones.

With a bit of patience, I could fix this. Just copy one file… wait for it to finish… then copy the next… Ugh! Who has the patience for that?

Instead, I wrote a script:

for ff in $(ls alloy-of-law-itty-bitty); do
  cp $ff /Volumes/OpenSwim/alloy-of-law 
  sleep 5

And… it worked! The tracks played in order.

Field test

During my ~40-minute swim this morning, I listened to The Alloy of Law by Brandon Sanderson the whole time. In that way, it was a great success!

However, I’m not sure I’ll continue to do this. I don’t usually do flip-turns, so there’s a moment at each end of the pool when the water drains from my ears. When that happens, the sound gets much quieter—water carries sound waves much better than air—and I miss a second or so of audio. I also found that I couldn’t quite devote enough attention to the book. Between maintaining my stroke form and tracking the other swimmers in my lane to stay out of their way, it was a lot of effort to follow the story.

There was also a small stutter when the track transitioned in the middle of the word. If I was planning to continue, I think that could be avoided by identifying gaps between sentences and splitting the track at that point.


I’ve heard people say that “coding is the new literacy”. I’m still skeptical of that idea. Literacy is a public good that makes all of society more efficient. Coding is useful sometimes. Still, I think this is a great example of applying programming skills to a problem in my own life.

Update: swimming-staging-area

I wrote a new script for copying the files across.

find ~/Music/swimming-staging-area -type f | sort | while read -d $'\n' ff; do
  relname=$(echo $ff | sed "s|^$HOME/Music/swimming-staging-area/||g")
  echo $relname
  mkdir -p "$(dirname /Volumes/OpenSwim/$relname)"
  cp "$ff" "/Volumes/OpenSwim/$relname"
  sleep 3

This new one:

  • Allows me to prepare all the songs and albums that will be on the headphones ahead of time, in swimming-staging-area
  • handles spaces in filenames
  • creates directories for the albums if they don’t already exist on the headphones.

With the new script, my workflow is to

  1. pick out albums and podcasts, putting them in swimming-staging-area.
  2. Rename the folders until they’re in the order I want.
  3. Delete the files in /Volumes/OpenSwim/.
  4. Run my script.

Update: shuffled folders

This is mostly working for me, with one small annoyance. I used spotify-dl to download a playlist of songs and I wasn’t careful about the filenames, so the songs are sorted by artist. I added a line to the script to shuffle the contents of any folder with “shuffle” in its name.

find ~/Music/swimming-staging-area -type d -depth 1 | sort | while read -d $'\n' dir; do
  sortFlags=$([[ $dir == *shuffle* ]] && echo "-R" || echo "")
  find $dir -type f | sort $sortFlags | while read -d $'\n' ff; do
    relname=$(echo $ff | sed "s|^$HOME/Music/swimming-staging-area/||g")
    echo $relname
    mkdir -p "$(dirname /Volumes/OpenSwim/$relname)"
    cp "$ff" "/Volumes/OpenSwim/$relname"
    sleep 3

Update: a kindred spirit

I heard from Christopher Woggon who has a very similar setup! His script adjusts the volume of each track to be the same, and uses the file-system sync instead of my sleep 3. Check out his blog post at https://christopherwoggon.com/blog/swimming-audiobook/