Waking Songs

Project Summary

I often wake with a song lingering in my head. These songs are not consciously chosen or triggered, and I rarely remember any dream context that might have been related to them. There is some overlap, but a large portion of them are not songs I ever get stuck in my head during the day. I decided to log all the songs playing in my head when I awoke, to the best of my ability.

Show me the data/insights!

• Up-to-date insights can be seen via Google Data Studio.
• I occasionally update a Spotify playlist with the songs that are available from their library.
• I am even less frequent about updating the YouTube playlist, but you can check it out here:

Not sure why you would, but if you want to make use of this personal data that I'm sharing with you, please contact me about it first. If you just want to mimic the project style, go right ahead! Would love it if you let me know. I'd love a glimpse into what happens in others' brains.

Method: data limitations and resolutions

Cognitive limitations

There are some obvious limits to my ability to log this data consistently.

• I may be too tired to remember to log.
• The information may disappear by the time I am conscious enough to write it down.
• I may not know the name of the song or be able to research it with what little is stuck in my head.
• The song may be a fiction created by my sleeping brain.
• I record whether I believe I heard the song recently, but this is fallible and the “recent” window may shift, as I am not good at pinpointing that sort of thing in time.

The first two difficulties sadly can’t be avoided, and there’s no way to know how much they cause under-logging, or whether there is a bias in the missing data.

Similarly the last item is unavoidable unless I cloister myself away from external music sources, and only use a music service that tracks exactly what I played when. And always remember to turn it off when I leave. None of those things are gonna happen.

If I can’t identify the song, I do my best to write down as much as I remember about the lyrics and style, and if I couldn’t make out any lyrics I ask SoundHound. If that doesn't work, I do my best to reproduce the melody line in a recording and ask someone who I believe is familiar with the music style. Usually I eventually manage to identify the song, but not always. There have been two where the artist and release date remain blank to this day, though I am still confident they were real songs.

Amazingly, so far I have always been sure whether the song was a real one or one that my unconscious mind constructed. For constructed songs I just record the current year as the “release” year, and “My Brain” as the author. There have been 9 of these made up “songs”, if you count the one that was a metal cover of “La Isla Bonita”.

Logistical sources of error

I write the data into Evernote shortly after I identify it sufficiently. The data in Evernote usually includes the artist, title, and whether I've heard the song recently. However, sometimes it contains only the title, or only a lyric that I believe will suffice to identify the song later. Once every month or three I go through my Evernote list and transfer all entries to a Google spreadsheet. I fill in the missing data based on Google searches. Unless otherwise noted, I fill in missing artists as the original or the one that comes to my mind by default when I think of the song. I listen to a recording to confirm whenever possible. I fill in the date based on the Google quick answer, or on Wikipedia or Amazon if there is no quick answer. Once every few iterations, I also add new data to the Spotify and YouTube lists.

I noted above the cognitive issues that can occur in the process of trying to make the original note. There are also some errors that can occur between that note and the data analysis.

• Any lyric I wrote down may not be sufficient to identify the song, though I haven't run into that problem yet.
• There can be an error transferring the data from Evernote to Google spreasheet.
• The data discovered in Google searches and especially Google quick answers (or Wikipedia or Amazon) is likely to be correct, but isn't infallible.
• The version that was stuck in my head may have been a different recording from the one I select when I transfer, possibly even a different artist. Checking the recording helps to reduce this, but isn't infallible.
• Occasionally my brain plays a mashup, and so far I have provided separate entries for the originals of each song with a note that they are a mashup, because I didn't anticipate this issue when I designed the spreadsheet. If you're a napper, you'd want to design the sheet so mashups were distinct from waking up multiple times per day.
• The transfer to Spotify and YouTube can add errors, compounding any prior errors. Also, many songs aren't available on one or both platforms. Fortunately, those lists are just there for flavor.

Future work

I intend to automate and improve the process, the data, and the visualization more and more over time. The initial process of writing down the song isn't improvable without some kind of scifi neural link, but when I get some time I can reduce the extra work I have to do to label, analyze (all automated, but can always be augmented!), and share the data.