A friend of mine recently had an usual request: He would like to have a way to listen to all of my piano recordings without manually switching tracks every couple of minutes. Imagine my shock when I learned that this poor dude is actually listening to them (??? no idea why).

It's not like I record a new piece every week, but I feel like I'd still be annoyed if I had to manually update this "master track" from time to time - so I automated that! I feel like this niche request doesn't warrant restructuring/modifying my website too much, so I opted to put this on a subdomain instead. You can find the new site at:

Again, it's literally just all of the piano pieces concatenated. I took the liberty to also do this for the guitar tracks, and also adding a third one, where all recordings are concatenated. I don't think I'll record new guitar stuff anytime soon, but I'm still working on my piano skills. Progress is pretty slow though, as this includes some pauses, dropping pieces, not really getting into others, and so on (lately I have been listening to way too much of Marasy8 arrangements, which are way outside of my skill level). Now I have a new metric to look forward to: Have at least as much piano stuff recorded as I did for the guitar :)

For future reference, I'll quickly cover how I can generate these master tracks with a bash script.

Technical Explanation

To be clear: Doing this by hand with e.g. Audacity is completely fine for a one-off. In fact, that was the first version I quickly put together before I looked into ways of automating it. For all things encoding related, ffmpeg comes to the rescue - although it wasn't really smooth sailing.

Turns out over the many years I recorded these things, I didn't use a consistent samplerate. That's not really too big of a deal, but if you just concatenate all of the audio samples into one big file, shit will break at some point. So I need to re-sample / re-encode the file. This hurts the perfectionist in me, but then again, it's literally just my recordings, that shit is not worth being stored losslessly to begin with.

Here's the strategy:

Gather all of the tracks in two directories (piano / guitar)
Re-encode all files with a consistent samplerate to FLAC, as an intermediate step (via FFMPEG)
Create lists of files to be concatenated (piano / guitar / all) (via find)
Use list to concatenate & encode the resulting file (via FFMPEG)
Cleanup temporary files
Upload the resulting files to the server (via rsync)

Here are some relevant snippets of the script:


# Convert Guitar Tracks
for file in $(find "$siteDir/recordings/guitar/" -type f); do
	name=$(basename $file)
	ffmpeg -hide_banner -loglevel error -y -i $file -osr 48000 "$tempDirGuitar/$name.flac" &
done

# ...


# Create lists
find $tempDirGuitar -type f -fprintf $listGuitar "file '%p'\n"

# ...

# Convert to single tracks
ffmpeg -hide_banner -loglevel error -f concat -safe 0 -y -i $listGuitar -acodec libvorbis -q:a 10 guitar.ogg &

The trick here is passing the correct arguments to FFMPEG. For concatenating files, it mostly supports a list provided by a text file in a weird format:


file 'myMusic.ogg'
file 'myOtherTrack.ogg'
file 'alsoThisTrack.ogg'

Flags like hide_banner or loglevel are just there to make the output quiet, while -safe 0 is required due to my file structure. I'm assuming it has to do with me using relative paths to navigate to my website holding the tracks, but I don't care enough to find out why exactly FFMPEG is careful in that instance.

Surprisingly, re-encoding all of that music, at least with my current track list, takes ~55 seconds. It would be well over two minutes if I didn't run most of the conversions in parallel (one big parallel block for all of the converting to FLAC, and then one parallel block for the three concatenated tracks).

Given that the "everything"-master-track is well over an hour long, it's a chonky boy, and probably not really suitable for something version controlled. The tracks of this site are in version control, so that should suffice in my opinion. That's why I'm just straight up uploading the files via rsync, instead of having some git repo with automated chron jobs and such.