We’re working on the world’s best music recommendation algorithm
The Muru Music team and I, are working on bringing you the world’s best music recommendation algorithm in the world — it will automagically select, mix and play a music playlist perfect for all your many musical needs, unlike other music apps that just keep playing songs in the same genre with the same pace or instrumentation.
We’re a tiny, seed stage mobile music startup in a space with very big, well-funded companies like Spotify, Apple Music, Pandora and YouTube. We are not their competition, but the value add they need to become profitable. I’m here to tell you how we are disrupting an entire market, using the fundamentals of DJing.
How can that be so? To understand, I need to hit the rewind button for just a moment.
I’m not a coder. For over 15 years I’ve worked as a professional DJ and music consultant. My entire career has been about finding and sharing new music with all types of audiences. More than 10 years ago I started studying Musicology to understand how people interact with music, how it affects our brains, the power of great playlists, and more interestingly, what makes music work when.
So what have I learned? We‘ve been approaching music analysis and classification all wrong.
I’ve been fascinated with the rise of music streaming in the past decade, especially to see what technologies have surfaced because of it. Thanks to companies like Pandora and Spotify pushing the envelope, a whole new layer of music technology has been built on top of those businesses which actually own the music. Australia has seen some home-grown successes too with companies like WeAreHunted (acquired by Twitter), and many startups that are breaking the mould, such as Hydric Media in Brisbane and GiggedIn in Sydney, all adding significant value to the music industry as a whole.
Listening to music was actually better when we had less of it
Remember how much each track mattered to you when your entire collection was maybe only a hundred songs?
As a kid, my musical choices ranged from Fela Kuti to Aretha Franklin, through Elvis back to Bach. Back then we took care of our collection and took our time when listening to music. We took our time when selecting an LP and we committed to the whole album. We really listened.
Music was much more personal, in the way we consumed it but also in the way we shared it.
Do you remember discussing music with your friends for hours? (in my case it would often lead to very intense discussions). You would spend time looking at the cover art, the credits and absorbing the lyrics — albums and singles were our prized possessions.
In high school, we would compare mixtapes we had worked on over the weekend. Mixtapes could make or break your social status so we took them seriously and could easily spend weeks on one tape, but it was oh, so rewarding!
I remember the first time I made a mixtape for a girl I liked — it took forever. I wanted to tell her how I felt through the music, but I also wanted her to love the mixtape so she would play it over and over again (and be reminded of me, of course). The amount of time, sweat and sometimes tears that went into these tapes were insane, considering it was for 60 minutes of music (equivalent to 20/25 songs.)
Those days have gone for most of us.
Now we live our lives at a faster pace, and mainly through our phones. Our attention span has significantly decreased and we are a lot less patient. While music has never been easier to access, we have somehow become more detached from it.
Such easy access to all music has had one major drawback — the grass is always greener, so people do not commit.
Today most people skip through songs, going for the instant fix rather than actually taking the time to actively listen. We have lost touch with something that is actually so natural to us.
When you used to have 50 vinyls at home, you had limited choice, but funnily enough it meant you got much more enjoyment out of those records and you literally played them to death.
These days, if you observe a teenager listening to music on their phone, you could easily mistake it for a game of candy crush. They are swiping and flicking through the songs as fast as humanly possible. I bet if you ask your teenager, they generally don’t know the name of the song, let alone how the song ends. (More on music related mobile behaviours by teens)
This type of consumption has even changed the way producers are arranging songs, going for the instant hook. In the case of music streaming, they want to get you to the music as fast as possible so you don’t get pissed off — in a way following the radio format of instant gratification.
We have stopped trusting our own taste and judgement in music and more importantly people have forgotten how music makes them feel. This isn’t just a trivial problem — it actually has a significant negative effect on our society.
Science has proven the tangible benefits of music in our lives — to fight stress, to heal, to relax and to stimulate our brains. “Healing At The Speed Of Sound” is a great book by Don Campbell and Alex Doman from Advancedbrain, which I highly recommend you read.
Why is music recommendation broken?
Most modern music apps and platforms are following the radio strategy of shoving music down your throat; music that you haven’t asked for and may not like. But it’s there and it’s easy. A lot of it is based on looking at big data, but we need to be very careful when we do that because… this is not 1’s and 0’s, it is music!
Ben Ratliff wrote a about it in his book “Every Song Ever: Twenty Ways to Listen to Music Now” More on that here.
“With online listening, what’s demanded of you in the moment of encounter is basically a non cognitive passivity”
Don’t get me wrong, there are great products out there and what Pandora, Spotify and others have achieved so far, is huge.
Spotify’s Discover weekly is impressive and I am a fan of Matthew Ogle, the man leading the team in Discovery. Matt was the founder of This Is My Jam, which was a very simple desktop app that let you search for a song on YouTube and post it to your friends and social network. It was incredibly simple but very effective. It made me stop and think and then encouraged me to reach out to my friends and nag them to listen to it.
Beats had a great idea with its on boarding process that asked a person what genres and artists they liked and in what situation they were listening to the music. This gave Beats a point of context but Beats never had a follow up question for when people didn’t like something. The feature was fun, but I was always disappointed with the results of recommendation across most genres.
So streaming services have done wonders in creating new music experiences, but even Discover Weekly has its drawbacks. If I spend a month listening to Ambient, because I need to concentrate, but then switch to Blues all of a sudden, I am royally screwed, because Discover Weekly will be filling my playlist with Ambient.
I have a user taste profile on Spotify which is then matched with similar profiles that have a similar “taste” — this is a great start to share new music with a user like me, but as I have no way of interacting with it other than skipping, or disliking, Spotify can make plenty of assumptions, but it can never really get the context behind the “why”.
Another approach is to use curated playlists. But I have yet to find a playlist where I like all the songs or want to listen to it more than once. And again, I have no influence over it.
So why don’t you just create your own playlists then? Well, let me remind you of the mixtape example — ain’t nobody got time for that!
So what are we left with?
We have access to all this music, but no really easy way to create an automated personalised experience.
This is where we need to take a closer look at DJs
When a DJ creates a set, they generally know what track they want to start with and a fairly good idea where they want that set to end up. The middle bit is highly dependent on the audience and how they are interacting with the music. This is where a DJ earns their stripes, by reading the crowd and understanding what their audience are after.
Somehow they just know the perfect track to play and they do this over and over again for a whole set…it’s almost magical. Listening to a DJ that can take you on a journey can bring you into an almost euphoric state — powerful stuff.
Pandora do a great job in finding songs that fit together, as they have hand curated their entire database through their Genome Project, but as much as I love it — the discovery is minimal and after I have listened to a particular station for a handful of times, it becomes very predictable and lacks the discovery element I desire.
No music app has quite nailed that yet, but I believe that the key to improved recommendation and discovery lies in how DJs and selectors create and more importantly adapt their sets to fit their audiences.
With 30 million song choices at any given time, we as users do need help to find what we love and in using these core DJ principles, I think we are onto something.
Instead of Thumbs Up/ Thumbs Down or skipping — why don’t we have a recommendation engine that can “read the crowd” and adapt the music journey accordingly?
Another problem with discovery in today’s recommendation engines is that they only stream a fraction of their entire catalogue. There are over 4 million songs on Spotify that have never been played. Cherie Hu, writes about it here.
I take pride in the playlists that I have created myself, but they are only useful to me on certain occasions — the remaining 99% of the time I want to just press play and automagically enjoy the perfect music appropriate for that moment. Don’t you?
So what if you could create “Presets” of the type of music you like, but that were dynamic and that you controlled. Presets that play the music you love and let you discover the gems you never knew existed — over and over again.
We need different music for when we workout to when we go to work, relax after a long day, have friends around for dinner… and so we need “playlists” that can adjust to these different occasions, while still playing the artists and songs we love.
Here is an example: I love listening to Deep House when I am working — it gets me in the zone and allows me to concentrate — but it has to be Deep House that is not too full on — then when I run I want the more up tempo Deep House that gets my ass into gear. Essentially two completely different playlists.
More importantly, I don’t want a random set of songs. I want a marriage of good tunes that flow. If I am listening to a song that I love that is mellow and smooth, followed by a song I have never heard of before, I expect it to have a similar vibe — a connection with the previous tracks and a connection with the following one as well.
If we currently dislike a song, we are played another song, but we have absolutely no idea why that was selected. Psychologically that puts up a barrier because we were not involved in the thought process.
So what if, rather than just disliking a song, you are asked — What don’t you like about it? There could be a variety of answers like — it’s too slow, I want something to dance too, I want to sing along, I want to hear the classics etc.
Now if a recommendation engine takes that into consideration when selecting the next track, all of a sudden we have a sense of ownership ... Why? Because we influenced it.
Music recommendation is a tricky field because in a sense we are dealing with human emotions and there is not one fixed magical solution that works for all.
We decided to create a recommender that answered all the questions I encountered over the years as a DJ and professional playlist maker. The result — Muru Music. Is it the right music app for you? That, you will have to answer for yourself.