Even though discovering new music has never been difficult, it is easier today than it has ever been in the past. However, music discovery is a personal issue: stumbling upon new music is remarkably easy, but stumbling upon new music that we will like and listen to several times for several weeks, months or years is notoriously hard.
With streaming music services, song stations are the new reconnaissance wings. Back when Rdio was around, I ended up with an entire playlist of about one hundred songs that I discovered solely through the service. Music discovery was top notch. I moved to Apple Music only months before Rdio shut down and was utterly disappointed with the poor song station feature. Despite the much-talked-about overhaul the app saw with iOS 10 last year, I remain just as disappointed.
Starting a music station from an instrumental-only playlist does little to ensure that suggestions are instrumental too. One would think this is a fundamental feature. Even if this were not the case, even if Apple Music decided to pay no attention to the song or songs I started the station with and if it, instead, chose to suggest songs with vocals, should the distribution of songs with and without vocals be even? In fact, besides the first song (or first two sometimes) nearly every other song blasts vocals, making it impossible for me to discover new instrumental-only music.
The old iTunes had a more granular system with star ratings from zero to five. Apple Music replaced that, for no reason at all, with the mysterious combination of ‘love’ and ‘dislike’.
The solution, more often than not, ends up being playlists. This is not ideal, but at least it is something. From what I hear, Spotify has a much, much better discovery service, so if Apple Music does not improve I will certainly consider moving to Spotify, but not before it is available worldwide.
That said, it is worth taking a look at just what parameters a song discovery algorithm must consider in the first place. At least on a fundamental level, it really is not something only Einstein can figure out, so why, despite the vast trove of data before them, the engineers at Apple have not got this right is beyond me.
Learning our tastes
The first is autocorrect. Apple Music has a search function straight out of the 90s. Typing in ‘J’y suis jamais allé’ with an english keyboard can leave you with errors that, understandably, may not be corrected. But what about typing in ‘Noah and the Whale’ with errors? A blank screen with no results.
Last.fm, once a growing name among music streaming services, famously introduced autocorrection into their product to help streamline metadata. Two people may spell something like Guns n’ roses differently, as Akarapat Charoenpanich of the London School of Economics pointed out in his study of data-based music discovery. It becomes important then to correct for this as well as to ensure the algorithm learns all supposed misspellings to help identify songs later on when a user searches for something, with or without mistakes in their search terms.
There is, besides robust search, the need to understand what songs someone likes. The old iTunes had a more granular system with star ratings from zero to five. Apple Music replaced that, for no reason at all, with the mysterious combination of ‘love’ and ‘dislike’. With that, you either liked a song or did not like it, but neither could you express nor could Apple Music know whether you liked one song more than another, a luxury afforded by a ratings system. (Apple eventually brought this feature back, but it remains a dummy that does not weigh in on the discovery algorithm.)
Social networks do not work
There is a semi-social feature called Connect that I was long convinced nobody bothered about. In the latest iteration of Apple Music, this feature was relegated from its spot at the centre of the service to a little section called ‘Connect posts’ buried deep below the ‘For you’ tab, farther down than anybody might scroll. Unless someone did scroll intentionally, though, the fact they reached the ‘Connect posts’ section without being distracted by any of the other suggested songs in the ‘For you’ tab speaks volumes about the quality of suggestions.
The whole idea of Connect should have been replaced by helping Apple Music integrate into existing social networks from the start. Encouraging people to share songs, letting others listen to song previews or, if they are subscribers too, letting them listen to the song, add it to their library etc. can help (but not much, as we will soon see). And then intelligently use this information, which says if someone added a song I shared, there is a fair chance we share our taste in music to some extent as well. This can then be used to compare the libraries of multiple listeners with potentially similar tastes to suggest songs to one from the libraries of the others and so on. Balanced well, this will lead to a ripple effect rather than an echo chamber.
That said, it turns out that social media is not a great help in terms of music discovery. For one, recall Mark Zuckerberg’s statement from back in 2011 when he was talking of Facebook and music streaming services working together: ‘You’re connecting the app and your timeline together, adding all of the activity and history in the app to your timeline, and keeping them in sync going forward’. Besides being creepy from a privacy perspective, it so happened that people got fed up of seeing music in their streams and Spotify led the pack in giving users an option to disable such sharing.
It is no wonder then that Mr Charoenpanich’s study concludes that a music service ‘cannot counter [the low value of social media] by improving music discovery features’ because sharing newly discovered music was what lowered the value of social media in what can only be described as a catch-22 situation.
My discovery of songs on Apple Music, that I liked enough to add to my playlists, has been infrequent. I used to add nearly every other song when on Rdio: discovery was top notch.
Social networks are therefore useful, but only marginally so. There are always a number of other metrics Apple can, though: there are some obvious ones like genre, singer, album or compilation, and there are less obvious but equally effective data such as the tempo of songs frequently listened to, the frequency of listening, connections between two artists (it is safe to assume a classical symphonist will rarely work with a rockstar) to suggest new artists and so on.
Some of my favourite discoveries
All of this is not to say I have never come across songs I like. I have, but they have almost all been vocals and my discovery of songs I have liked enough to add to my playlists for regular listening has been more infrequent than during my time with Rdio. In fact, I remember I used to add nearly every other song when on Rdio and discovery was top notch.
Among the ones I have come across, below are some I have liked (again, all vocals, no instrumentals). They may not all be new to you, but they were new to me and if you have not listened to these before, I certainly recommend that you look these musicians up. Below is a list by artist with select songs mentioned to help you get started.
- Nick Mulvey. An English musician who studied in Cuba and (perhaps I should say therefore) whose music has a sort of laid back feel. Start with ‘The Trellis’.
- Laura Marling. Another English musician, this time from the folk community, with soothing tones and flowing music. One of my favourites is ‘Blackberry Stone’, another is ‘Alpha Shallows’.
- Alexandra ‘Shura’ Denton. This young woman has a fast-growing fanbase. Also, I now realise that I have far too many English musicians on this list, unintentionally. Unlike the previous two, Shura’s songs are faster, with more of an electronic and pop vibe. ‘Touch’ is famous among most but I like her subtler ones like ‘Make it up’ and the catchy ‘2Shy’ (sic.)
- Benjamin Francis Leftwich. Another English folk singer with some beautiful guitar and equally good words. Listen to ‘Shine’, which Spotify called its most addictive song in 2014. My favourite is ‘Maps’.
- Marika Hackman. The rhythms in her songs are different. That is all I can say to describe them. But they are interesting and melodious and a bit lose on harmony with the accompanying music in a strangely alluring way. Try the popular ‘Animal fear’, but listen to my favourite, ‘Mountain spines’, before that.
- Lucy Rose Parton. There are few singers you can describe as having a voice that feels like it is being sung directly into your ear. Lucy Rose is one of them; another I can think of is George Ezra, but he eventually got far too loud and monotonous for my taste. Lucy Rose is more thoughtfully quite. Try ‘Into the wild’.
- Rachelle ‘Rae’ Morris. It appears this list is destined to carry only English singers. Rae Morris is a pop musician whose début album, Unguarded, did not become unusually famous, but is still pretty good. I recommend her song, ‘Cold’, sung with Fryars.
- Genesis. A band with Phil Collins in it. If that is not reason enough to give them a try, pick up their song ‘Hold on my heart’. Interestingly, they are the oldest musicians on my list, having performed from the 60s–80s, paused briefly and picked up again in 2006. Although it is from 1991, I only discovered ‘Hold on my heart’ recently.
- Brigitte. A French duo I could not help but toss into this list because nearly all their songs are lovely. Made up of Sylvie Hoarau and Aurélie Saada, the band is supposed to stand for ‘the woman in plural’. Listen to ‘À bouche que veux-tu’ and ‘Battez-vous’, but you probably will not be able to stay away from their other songs anyway.
- Noah and the whale. Perhaps among the better known on this list, alongside the previous three, this was a short-lived English rock band that played from 2006 to 2015 and left quite an impression on its listeners. Start with ‘Last night on earth’ and ‘Heart of nowhere’.
Apple thinks my music taste is best described as ‘English talent only’. Jokes apart, it took me six months of song stations to settle on these because recommendations were poor. Apple Music as a service is good, but it can be much better, and I hope iOS 11 is the answer. ❖