Anki for Music — one year later
A year ago, I wrote this story about using Anki flashcard software to manage the catalog of music that I was learning and maintaining. At the moment that I wrote that piece, I was one week into learning about Anki, and now it’s been a year, so I thought I’d revisit that story and see how things have changed. It’s really helpful to have reference points like this; it helps me gauge my progress.
One thing that I already knew when I wrote that piece: using the flashcards immediately relieved anxiety for me, because it told me what song to play next, so I could just keep playing, and very quickly my playing became almost continuous. I could play a song, flip the card, play the next song — if I played for an hour, I got probably 58 solid minutes of playing out of that hour. Right away, I was getting a lot more bang for my buck with regard to time.
But tools like Anki also make it really easy for you to obsess, so the trick is getting the value out of the tool without becoming a slave to it. It took about six months for me to hit equilibrium; for the first six months, I never ever finished a day without a backlog. The backlog was, of course, my fault, not Anki’s — it sounds ridiculously simple when you put it this way, but if you are maintaining a couple of hundred pieces of music, you can’t play all of them every day. You just can’t. But when I play a piece and there’s something I know I need to work on, I really want to have it come up again the next day. So the first skill I had to get better at was deciding how long I was willing to wait before I played it again. I had to fight myself on it a lot — I still do, but I’m getting better. Once I hit that equilibrium, I really wanted to maintain it, because Custom Study is a lot more flexible, so I made staying even a priority. For the past 6 months, I have maintained that equilibrium even through illnesses and vacations. It’s a big achievement, and every once in a while, I pause to congratulate myself.
Along the way, my settings have changed a lot. And I’m using sub-decks.
So this is how it looks at the end of a workday. My “music” deck shows 85 cards due, but 81 of them are from the “marieNme” deck. MarieNMe is a duo that I’m in, and my partner in that duo is out of the country, so those 81 cards will be due until he gets back in April. Day to day, I focus on the “marie” deck and its sub-decks.
The reason 4 cards are already due again is because they’re the new ones, and I have them on a 4-hour rotation, so they fell due again after I stopped working. The “Custom Study Session” deck is made up of cards that are due over the next 3 days, so if I am able to work from that deck every day, I stay current and have no backlog.
It says I worked on 40 songs today, in 201 minutes. This is turning out to be the normal workload, three to four hours, thirty to fifty songs, depending.
The deck options for that “music” deck took a lot of playing with, but I am currently very happy with them. The Anki user manual is very good, but it took me almost the whole year to understand most of what it was trying to tell me. Here is the way I have things set:
This is the New Cards options screen for my “music” decks. The learning steps are set to 4 hours, then 8 hours, then 8 again, then a whole day. This is just for songs that I’m learning. Once they become “learned”, they will use the “review” settings on the next screen. Last year, I had the initial interval set to 120 (2 hours), but that turned out to be crazy — I would get a song again in the same sitting. So I shifted it to four hours (240), so that if I sit down again later in the day, I play the song again. If you do that with a song for a couple of days, you learn it pretty fast and it graduates to “review” status.
These are the “review” settings. When I started, I had a maximum interval of 8 days — that was madness. I have several hundred songs in this collection, and as the year went on, that interval got longer and longer. It’s now sitting at 6 weeks, and that seems great. Songs that I really know well don’t need to be played more often than that to be remembered. If I get bogged down, that “maximum interval” is one of the first things I look at changing, especially since I can always send a card back into rotation if I start to forget it.
This is the screen where I control what happens when I send a song back into rotation. Leaving the steps blank causes it to figure a new interval with the information it has. So, for example, if I had a song set to 30 days, but I don’t remember it very well, it’ll send it back in like 3 days, and see how I do from there. The “leech threshold” is set high, and the action to “tag only”, because I don’t really care about leeches. It’s tracking how many times I restart the process, but at this point that’s not something I care about; I just want to make sure I know the songs, so I send them back as many times as I want.
Letting Anki manage my workload has been life-changing for me. I am prone to anxiety, and one of the biggest triggers of anxiety for me is uncertainty. I am letting the tool be responsible for the millions of little decisions that you make while you’re working, and it frees me to look at the bigger picture. I’m not focusing on deliverables of any sort; I am trusting that doing the work will take me where I want to go. And pretty much, it’s working.
When I started this, I had a soft goal of 50 songs a day, and that is what it has worked out to. If a song comes up that I don’t really feel like doing, I have learned to remember that there was a reason I put that song in the collection to begin with, and I am learning to just trust that I should do the song — it becomes less work to just do the song than to argue with myself to be lazy. And so I do more work, and I get better, and everything becomes less work to do — it’s been a wonderful thing.
I am also a programmer, so I quickly saw how the tagging feature that Anki has built into it could be used for meta-data. I started tagging songs by artist and the year they were written, and soon I had a web of interconnections between the songs. Anki’s browser wasn’t built to handle the kind of cross-referencing I wanted to do, so I decided to write something to handle that for me. Anki is free and open-source, and its database is in SQLite. With a little digging, I found the database schema on github. I use Python to extract the data I want into a JSON file, and am now building software to manage it.
So now, in addition to spending 3 to 4 hours a day playing the music (and generating more Anki data with every session), I’m spending another couple of hours coding around the data I’m generating. Interestingly, even though I use Anki to enter the data (so I avoid having to write and maintain a multi-platform data entry system — Anki just does that for me), the data that I end up extracting from it has absolutely nothing to do with Anki at all. Anki’s database could tell me how long I spent working on each song, the dates and times, etc., but none of that data gets extracted — the data I’m interested in is the list of songs that I can do from, say, 1983 — and how many of them were top ten hits, and what artists collaborated to produce them, things like that. It’s a wonderful adventure, and I’m hoping that next year at this time, I’ll be writing a piece that’s more about that aspect of my work.
But for now, I’m just reveling in how much improved my playing is, my singing is, my understanding of music is, and how much using Anki has helped me to facilitate that growth.
Thanks for reading!
