You Have Not Yet Heard Your Favorite Song: How Streaming Changes Music

by Glenn McDonald

From Spotify’s ‘Data Alchemist’ comes a comprehensive guide to music’s digital revolution.

Starting with the 1990s days of vinyl and CDs and moving quickly to the explosion of new technology in the 2000s, and then onto the rapidly changing soundscape of the 2020s, YOU HAVE NOT YET HEARD YOUR FAVOURITE SONG is a fast-paced, factual look at:

• How streaming has changed the musical landscape for all global music genres, from rap to punk to jazz

• How people worldwide listen differently and what type of songs and artists are popular in each country

• How music gets onto the streaming platforms, and what platforms like Spotify know about you as a listener

• How Spotify and rivals reward artists and the record companies and the lengths some musicians go to to game the algorithm

• How to exploit the fact that pretty much every song ever recorded is available now, at the touch of the button – and discover new music. (Statistically, you have not yet heard your favourite song...).

With 10 free exploratory playlists downloadable by QR code

Genres: Music, Nonfiction, Technology, Business, Cultural, Art, Design

320 pages, ebook

First published June 20, 2024

About the author

Librarian Note: There is more than one author in the GoodReads database with this name. See this thread for more information.

Reviews

[Christopher Dalla Riva · Rating 5 out of 5]

I work in the music industry. I make music. I also write a somewhat popular newsletter about music. If I had to tell someone to read one book to understand how music works in the era of streaming, I would probably recommend this book. Written by former Spotify data guru Glenn McDonald, this book covers everything from how we define genres to algorithmic recommendations to payment schemes. It’s a book that I wish I could have written.

[m. · Rating 5 out of 5]

     Ця книжка не лише відповідає на питання «як працюють стрімінги?», а й «чому вони блядь не працюють так як треба і чому спотіфай мені рекомендує йобану русню?»

     Рекомендував би її прочитати усім, для кого музика – це щось більше ніж кнопочка плей.
     Книга не так відкриває якісь нові знання, як радше впорядковує усі ті краплини, про які ми й так здогадувалися.
Та це напевно та книга, яку захочеться раз в декілька років переглядати. + попідкреслював собі пів книжки цитат, щоб застосовувати їх як аргументи у Інтернет-срачах.

     Ґленн, як колишній дата алхімік (технічним людям треба якось вже бути простіше з назвами своїх позицій в компаніях, нє?) Спотіфаю максимально доступно та задротсько (в хорошому розумінні) розповідає про індустрію музики і її розвиток від фізичного медіа до власне стрімінгів.
     Як змінилося сприйняття музики, чому музика альбомами відмирає, скільки музики фізично можливо прослухати взагалі, як найправильніше розподілювати доходи, що сервіси знають про нас, чому ви ще не почули свою улюблену пісню та чому алгоритми ніколи не помиляються.
І чому на всі ці питання немає однозначної відповіді.

     P.S. Рекомендує русню – бо слухають русню ¯\_(ツ)_/¯
     P.P.S. Навіть якщо русню не слухаєте ви особисто, саме так.

[Ted · Rating 5 out of 5]

To be honest, I'm a bigger fan of what Glenn McDonald has brought to music than I am of any particular musical artist. Through his work at The Echo Nest, Spotify, and (especially) Every Noise at Once, I've experienced so much joy and discovery. I've also loved his writing at furia.com, where he posts new-release poetry, explorations of music genres as communities of love, and (increasingly rare) bloggy bits. His Twitter feed was a delightful, patient (and occasionally sardonic) drumbeat explaining that common assumptions about streaming economics are, in fact, completely wrong. So I was very excited to read McDonald's thoughts at book length. If you've ever wondered about those strange genre names in your Spotify Wrapped, wondered how streaming companies pay artists, and wondered how to explore the vast universe of (almost) all the world's music available all at once, I highly recommend this book. If you just enjoy listening to music, I highly recommend this book. If "how streaming changes music" sounds a little too specific of a topic for your book reading, may I suggest that the secret subtitle is "how streaming changes joy." And if you're still unconvinced, go check out Every Noise at Once, marvel, and then pick this book up.

[Sebastian Gebski · Rating 3 out of 5]

Half of this book is perfect (literally), the other half is ... far from that. Sadly, the 2nd one is much more important ... What do I mean by that?

First of all - it's a book by a former data engineer in Spotify. It's not a book about Spotify, but the recent changes in the music industry - caused mostly by the streaming revolution. It has been written as a collection of essays. Each one is pretty much a long answer to a very specific (& relevant!) question. This is the perfect part - the questions (chapter topics) are absolute bullseye. The author has nearly perfectly captured the questions and speculations regarding the music industry, e.g.:
- how does compensating in music streaming look alike?
- are there alternative models? would they be more fair?
- how do music streaming service profile you (what are the options)? how do they personalize the content for you?
- does streaming models create a new monopoly/oligopoly?
- is it possible to "cheat the system" when it comes to compensation for an artist?
- how to analyze and classify music?
- we don't own the content anymore - is it a problem?
- how our tastes and the general music market evolve?
etc.

These are just few, but the book is literally FULL of good, relevant questions - unfortunately, it's the answers that are the issue ... It starts quite nice, I enjoyed the basic intro to streaming economics (even if I was already quite familiar with how it works), but the further into the book the worse. The answers become very philosophical or (e.g., in the Part 4) they are pretty much a "word soup" I couldn't learn anything from ;( I didn't expect to learn many "secrets of the kitchen" (the author stated clearly he has no intent to reveal any confidential practices/projects), but I expect some proper, in-depth analysis. I don't think I got it ;/

That's why it's only 3.1 stars. Because the book is a good starting point, but you need to do a lot of work yourself (when it comes to research).

[Kathryn Carlson · Rating 5 out of 5]

“What machine learning most enthusiastically facilitates is turning cultural problems into technical problems, reducing the need for qualitative human input, so that they can be addressed by engineers who don’t have to know or care about the cultural details.”

written by Glenn McDonald, former Spotify data alchemist and creator of my favorite website everynoise.com. also the creator of the genre classification system on spotify (yes he coined the term escape room)

a great read for anyone who is fascinated by genre, music exploration, and personalization algorithms. McDonald discusses the responsibilities of the humans who create algorithms and paths for moving forward in this burgeoning field

“demand that algorithms both earn your trust and maintain it”

“Love what you love, which needn’t be only music. But pay attention to how things absorb your time and attention out of inertia instead of love.”

…

also of note: the Christmas music anomaly in data that has to be mathed out of algorithms, and McDonald’s fervent hatred of jazz 😂

[Stephen · Rating 5 out of 5]

While this book bears all the signs of a blog smooshed into book shape, it is saved by being pretty much right about everything.

I'm one of the few people in the world that worked on automated music recommendations professionally. It manages to articulate the view from the inside in an accurate way, which despite wider histrionics, has plenty of intriguing areas of optimism. I particularly liked sections on how to think about genres, the relation between albums and playlists, and the unusual ways gen z can form musical communities. My biggest criticism is his actual taste in music, for which there is famously no accounting

[Ben Schulman · Rating 2 out of 5]

The beginning and middle sections of this book contained some really interesting information about the music industry and how artists get paid. The latter half of the book felt like the author listing his favorite songs and genres. It felt like the interesting and informative content of the book could have been boiled down to an article or blog post.

[Giray · Rating 3 out of 5]

Out of respect for what Glenn has done (Spotify discovery algorithms, everynoise.com, decades of blogging to name a few examples) to improve the experience of listening to music, this book deserves a read. However, I would have appreciated the content more in blog format - each chapter felt only slightly different and I didn’t sense the progression, so by the end, my brain wasn’t even registering new ideas anymore. He does like Turkish rap!

[Dmitrii Petrov · Rating 3 out of 5]

This book could have been a long-read article. And it would be way better.

I still recommend this book for people who are interested in how streaming works and affects our listening. Or just to explore how some people can be very curious about over-analysing data. It's quite fascinating.

In other cases, this book could be boring and useless

[anchi · Rating 4 out of 5]

我覺得書名「串流音樂為何能精準推薦『你可能喜歡』」有點誤導，因為本書實際上談論了串流音樂的許多面向，包括演算法、經營方式、以及音樂背後的各種秘密。從書中可以看見，作者是個對音樂非常非常有熱情的人，我也從書中學到不少有趣的音樂類別。我只給四顆星的原因是，書裡提及的內容沒有想像中的深入，比起認識不同類型的音樂，我比較想知道更多關於演算法的秘密。

[Connor · Rating 5 out of 5]

More a collection of musings and short essays but it gave me a lot to think about in regards to music and the form we consume it today, and taught me even more about how underappreciated a tool Spotify is for learning more about our world than we ever could alone. There's a good amount of personal philosophy-sharing tangents that I think I'd be more annoyed about if McDonald didn't seem like such a cool guy lol

Big recommend!!!!

Fav quote from near the end "We can form ourselves into chords but only if we sing out loud". Wasn't expecting a lot of philosophical positing of the importance of celebrating and loving openly and widely but happy I got it

[Rónán Ó · Rating 4 out of 5]

Cool insights into the music landscape. The writer is clearly very passionate about music and knows a lot about using data. Would have preferred if it went a bit more indepth at parts but happy i read it, even though I'll probably never look at it again

[Daniel Rodríguez · Rating 4 out of 5]

Great book 😍, what I like is that it was written by a software engineer, Glenn explains the technical side of the Spotify's business model and how Glenn's passion for music motivated him to create amazing pieces of software that millions of users Jan enjoy today. Another great aspect is that Glenn used loads of music references, he recommended plenty of bands that I didn't know before and he talks a lot about metal music 🎸🤘

[cehryl · Rating 5 out of 5]

incredibly insightful. closed the book with great answers (transparent view of how streaming and algorithms work for artists and consumers) but more importantly, even better questions. thank you glenn for writing this book and for existing and sharing your knowledge and perceptions and for championing music. i think everyone who likes music should read this.

[Anthe · Rating 3 out of 5]

I liked the technical explanations (even though they were not always as in depth as I'd like) of how the different mechanisms in Spotify work. I think those were also explained in a quite accessible manner. Furthermore, I applaud the attempt at a sociological analysis of streaming, but I don't think the author is reality qualified for it. Especially the latter half of the book went off the deep end.

Sometimes the book read as a hippie manifesto in favour of streaming services. I was never really convinced.

[PJ · Rating 5 out of 5]

Borrowed this book from the library and it renewed my love of music. It was so fun to read about all these genres I wasn’t aware of, and I thought I already know quite a lot. I kept pausing my reading to check out the music on my streaming provider (well, not Spotify).
So when I returned the book to the library, I bought a copy just to keep.
Thank you, this book made me happy.

[Accio_sunshine · Rating 4 out of 5]

Started off with why my Spotify wrapped is meh this year and discovered this book.
Glenn writes with so much passion that I cannot wait to discover my next favourite song; beyond language and culture :))

[Sofie · Rating 4 out of 5]

"the world is more full of music than your heart is, yet, but the more of our shared world you hear, the more your own world expands, and thus the more joy you can fit inside of yourself, until you can't stop it from spilling out again constantly as you walk, as you sing along, as you breathe."

[Jessica Brown · Rating 4 out of 5]

This book made me what to explore music with a wide open mind. I'd say it achieved its purpose.

[Peter Bergmann · Rating 4 out of 5]

A passionate defense of music and good look inside algorithms. Interesting read for anyone interested in how data shapes our lives or is just a music nerd.

[Annie Song · Rating 5 out of 5]

Probably a 4.5, but as a Spotify super user who studied computer science, I’m probably the ideal reader for this book

[Gavin · Rating 3 out of 5]

you work on technology, or it works on you

Rare insights into the current situation by an ex-insider and massive fan. (McDonald got canned from Spotify in 2023 after ten years, years which spanned many huge shifts, including towards the automated playlists. He played a large role - he has in a narrow sense thought about music more than almost anyone; the sense being musical epidemiology, music-as-digital-information.)

He's not recanting; he still likes streaming and believes in the 2013 era "data storytelling" thing and gives an actual argument for what we've done to ourselves. Music used to be shopping (money-bound) and high-risk; now it's time-bound exploration. The nerds have less power now:

In general the economic history of music has been a movement away from royal patronage towards populism, from court orchestras to public performance to recordings. The gap between the most powerful and the least tends to get narrower.

But streaming is an amazing deal for those same nerds (2 albums a week was $1500 a year, now n albums a week is $120 a year for any n), but this is subsidised by an increase in spending by the disengaged ("average per-person music-spending in the LP/CD age put it somewhere between $25 and $60 per year").
At no point does he reflect that more music might have downsides - he lacks the distinction between (deep) listening and (shallow) hearing. I like many albums of the last ten years but I only have a deep, 00s style connection with four of them. I don't think this is a property of the music, I think it's because of what the new platforms have made me: distracted and impatient. He does say this:

The cultural result of this control, however, was a shared experience of music. As a listener you only knew the hits, because only a few things could ever become hits, but this was true of everybody, and thus everybody who knew music knew the same music. Or as a listener, at least, you probably felt like everybody who knew music knew the same music as you.

(Nope!)

This is what we mean by ‘the monoculture’ when we talk about the history of music. But there was music everywhere, and it mostly didn’t escape the place where it was created, so there were actually many monocultures scattered over the planet.

Pro-Rata vs User-Centric Payments and the Microeconomic Implications of Hypothetical Notions of Fairness in Streaming-Royalty Allocation Schemes‌‌‌‌‌‌‌: f more-active listeners listen to more-popular artists, then the pro rata scheme will be regressive in the economic (and social) sense, compared to the user-centric one, taxing the less-popular artists to consolidate wealth in the most-popular. Whereas if more-active listeners tend to listen to less-popular artists, then the pro rata scheme will be progressive, redistributing money from more-popular artists to less-popular ones. Thus this very basic question can be answered quantitatively by computing the average number of monthly streams per Spotify listener, and then calculating the average monthly total plays per artist of the artists played by listeners with fewer streams than that vs the ones with more. A ratio greater than 1.0 shows that pro rata is regressive, a ratio less than 1.0 shows that it’s progressive, and either way it does so without revealing any remotely-proprietary actual counts or averages. When I monitored it at Spotify, it hovered around 0.83. More-active-than-average listeners play artists who are, on average, 0.83x as popular... Pro rata is progressive, and user-centric would be regressive.

there are only four major listening modes where playlists are not ultimately the most-traveled path to albums... background noise, where the whole point of a rain-sound playlist, for instance, is to loop in the background without you paying attention to tracks... compilations from genre communities specifically built around individual contributions from many participants... very old genres that only have a streaming audience at all due to nostalgia, and nostalgia exaggerates selectivity... you’re nostalgic for hits you haven’t forgotten and probably can’t... Electronic Dance Music... a function of DJ mixes as the native form of dance music

In the back half he gets political, which is nice.

Streaming is surveillance capitalism... A streaming service [has to surveil] logistically and legally, because it has to pay royalties for the music, and those royalties depend on the song, the account, the date and the time... They know what we play, but they can’t see whether we’re dancing enraptured while the music spins, or two rooms away folding distractingly crinkly laundry. They know which songs we put on playlists, but not what the playlists are for. They know when we play a song ten times, but not whether we’re doing it because of the drums or in spite of the banjo. They know what we click on when we search, but not whether we knew what we were looking for. They can’t tell whether we scroll past their offerings in disdain or distraction. They can’t distinguish between joy and irony... You’re not really being surveilled, but that’s not due to teleological humility or ethical judiciousness, it’s just that invasive surveillance doesn’t address any business problem that isn’t at once more easily and more effectively solved by regression to the mean

Statistical correlations aren’t value-statements, though, and nobody was saying that those 135 manly country songs were more important or more culturally relevant to Country Music than Carrie Underwood or Martina McBride or Sara Evans.
Or at least that would be true, arguably, if it weren’t for that big word ‘Recommended’ at the top. You can’t blame math for multiplying, but you can blame people for any time we label the results of math with emotional implications. We could have said ‘Songs most often found on playlists with this title’. But we didn’t. We wrote ‘Recommended’. We attributed value to popularity.

"Peaceful Piano", which is a burbling font of gently directionless soft-focus piano music... it quickly created a shadow mini-economy of pseudonymous filler purpose-made to appear there or in a spiraling array of imitators. Spotify has been accused of making some of this music itself to avoid paying royalties, which it doesn’t [ED: they do], but this is a meager legal consolation... Meanwhile, pseudonyms prompt many other questions. Are there people behind these aliases? Are the same people behind many of them? Is this how the AI apocalypse begins? ...similar aspirational waves of ambient guitar, dinner jazz, music-box translations of pop songs for babies, or music ostensibly intended to calm your pets while you’re away. Situational repackagings of resolutely unthreatening music now comfortably outnumber those of background noise.

* Very weird that this book by a famous data scientist has no graphs in it, leading to silly passages like

In raw dollars, US ‘recorded music revenues’ peaked around $14.6b in 1999, dove to about $7b by 2015 during the crash, and built back to $15b based on streaming by 2021. That seems great until you realize that $14.6b in 1999 dollars is more like $23.7b in 2021 dollars. The crash actually cut the industry to about a third of its size, not just a half, and 2021 is effectively level with 1991 and 2006, not the 1999 peak. Or 1977-1979, for that matter. So if it feels to you like things in the music business are a lot worse than they were, that’s because they mostly are.
But they’re recovering. Streaming took seven years to go from $1.2b (2014) to $11.5b (2021) in adjusted dollars. CDs were at $1b in 2021 dollars in 1985, and by 1992 had only reached $10.4b. Paid downloads were at $1b in 2006 and by 2013 were already falling from their 2012 peak of $3.4b.
And if you’re participating in any part of this, you are part of music’s recovery, and part of its potential future. If you can afford to subscribe, you should subscribe

* His quick history of the pre-internet music completely ignores the existence of DIY labels, college radio, and pirate radio. OK: 98% of people paid no attention to this option or this music. But still the internet was not a step change in liberty for some of the most influential music ever (punk, hip-hop, house, lo-fi...). Rick Rubin was DIY!

* Surprising view:

Machines don’t learn any more than they know, so when a human tells you a machine is learning, you can safely infer that nobody is learning. Sometimes ML works despite this, and sometimes it causes chaos. So the good news is that the robots are not plotting your downfall. The robots have no domination plan, and indeed no plans at all, and anyway there are no robots.

On a reliabilist, functionalist, Hayekian, or externalist account of knowledge, they obviously know things. McDonald might have an internalist or virtue-epistemic view but more likely he has no view. There are accounts of learning which don't boil down to improving representations after seeing data, but I don't see why I should respect them.

* Endearing that he calls himself a music "'artist'", sceptical as I am of the recording industry's standard honorific.

* Ah that's why people like that band:

Glass Animals’ ‘Heat Waves’, featured in a Dream-related romantic fan-fiction, and snowballed from there to an eventual Billboard #1 in its 59th chart-week, which was a record at the time for the slowest rise to the top. This is what a community looks like, of creators and their audience and their shared interests.

* This is a lovely para:

A physical album release required lead-time from manufacturing plants, art-direction, typesetting, printing, trucks to drive boxes of things from place to place, hands putting things onto store shelves, buyers going to stores, parking, unpackaging, playback machines with dizzily moving parts. An artistic album, similarly, required interleaved critical masses of conception, composition, personnel, performance, recording, production and sequencing. And these dual accumulations of investment didn’t technically require, but firmly suggested, the other slope of an album cycle... expensive things need to seem serious, and serious things are hard.

*

Saying you’re a Machine Learning Engineer is like saying you’re going to microwave anything you’re given to cook...

The old algorithm’s set had been carefully filtered in many obsessive ways over the years as we learned which kinds of songs work well in algorithmically generated playlists and which don’t. Songs with long silent gaps are bad, and we have software for detecting gaps. Commentary tracks detached from the songs they’re explaining are bad, and we have software for distinguishing speaking from singing. Kids songs and Christmas songs are bad to mix with non-kids songs or non-Christmas songs, and we have things for those, too. Very long or very short songs are bad.

* Ah, that's why Discover Weekly is so good - it's reusing human work

Discover Weekly is based on co-occurrence of songs between listener-made playlists. And instead of using simple math... it uses... vector embedding, which turns each thing (in this case a song) into a list of numbers.

He sometimes betrays some of the philistine energy you would (wrongly) expect from the wizard behind the curtain of the echo nest - he likes schlager, he hates jazz -

I listen almost exclusively forward in time, to new music, so you might think that the fate of old music would be of negligible consequence to me. If I hold my head at the right angle, so the right nerves in my neck are compressed, I can almost believe that old songs don’t have to survive, because we can always make more, and more realistically can’t stop making more. But no. New songs exist in historical context...

Genres don’t really exist, they’re just words people use to talk about things... [but, enlightened:] a genre is really a community

But overall a nice inside peek, mildly critical, but it badly needed another copy-edit to take out the mawk. See also Pelly and his fairly good blog.

[Isaac · Rating 5 out of 5]

I do not typically read nonfiction books, but as a music lover, a genre explorer, and someone who too late discovered the project Every Noise at Once, this book seemed promising. It turned out that 'Promising' was an understatement. It altered my perception of the streaming music industry; gave me easily understood reasoning behind the logic that much of streaming runs on; provided wonderful analysis, criticism, breakdowns, and proposed solutions and ethical guidelines for the morality of algorithms; opened my eyes to the immense diversity of music that exists; provided me with new avenues for discovery within this diversity of music; did all of this while being extraordinarily well written, making me laugh, and using uniquely descriptive metaphor to illustrate concepts; and made me use semicolons for this list, which is maybe the third time I've ever needed to do so in my writing experience.

If you have a love of music, an interest in music discovery, and a background in computer science, you should read this book. If you don't have a background in computer science, you should also read this book - none of its text included over-the-top technical descriptions, and any technical descriptions were done in a very down-to-earth, practical way.

This is likely the first nonfiction book I have read since college, and it is also one of the best books I have read since college. I cannot wait to hand it off to any friends who will consider reading it; I only wish I had more copies to hand off to more friends.

[Stijn · Rating 2 out of 5]

Quite uneven. Hard to say what was wrong, somehow technical but also not.

[Aram · Rating 5 out of 5]

One of the easiest books to read about algorithms you'll find, chock full of insights about how music, the music industry and the world of streaming works. Perhaps the best part is that the casual fun writing is shot through with the author's music suggestions and humanity, giving the whole piece a warm friendly feel. It's also filled with potent quotes about the future we should want to build with algorithms, computers, and art. Every Spotify user should read it. Then everyone else.

[Eskil · Rating 4 out of 5]

Fascinating book! I still don't understand the payment model or how most artists make literally no money at all, but the music recommendation stuff was great reading. I think something in the business of music needs to change but IDK how or what.

[Ecem Ünal · Rating 4 out of 5]

Love the idea behind Every Noise at Once, so this was a very fun read. Was interesting to have a peek into the streaming industry and thoughts that go into the discovery alghorithms. Secretly enjoyed the running theme of symphonic metal throughout the book as a fellow metal fan, but the mention of interesting and niche genres inspired me to broaden my horizon as well.

[Dave · Rating 4 out of 5]

A fun and interesting read, even when it resembles a lecture or TED talk that I don't especially agree with

[Danny B · Rating 2 out of 5]

This book is actually very informative and describes many aspects of streaming and data that are beneficial and interesting to know. I thought as a creative I would’ve been more drawn to this book, but I think the way it was written went off topic a bit too much and was either too wordy or written in a complex way that didn’t necessarily make too much sense. It was interesting enough but I couldn’t finish it to be honest without it feeling like a chore.

[Jeremy Walton · Rating 5 out of 5]

Stream a little stream
I bought this because the author sounded as if he had an stimulating combination of interests: passionate about music and discovery of new sounds (on p255, he gives one of his principles as "if you have the choice between listening to music while you do a thing that allows you to listen to music at the same time, and a thing that doesn't allow it, choose the thing that does"), plus the technical skill to mine the listening data gathered by a streaming service (specifically, Spotify, whom he worked for until 2023) in order to generate suggestions for each listener about what (new) music they would enjoy.

He gives a good account of the way Spotify works - particularly the way it allocates the money it earns from listeners and advertisers each month by dividing it up according to the total streaming (by all users) which happened that month. He argues that this is biased towards rewarding more obscure artists than the mainstream ones, because listeners who like the former are likely to listen to more music than those who are focussed on (say) Ed Sheeran. He's a passionate advocate for streaming (as opposed to downloading, or purchasing CDs) as a means of accessing (just about) every song ever. This is the origin of his snappy title; it's not spelt out explicitly, but assuming there's an infinite amount of music still to be heard, then it's likely that a better song than your current favourite is in there somewhere (a suggestion that might not take into account the difference between quantity and quality).

From a technical point of view, he points out how streaming is much better suited for the analysis of listening patterns than downloading or CD purchasing, which he memorably likens to "trying to understand social trends in body weight if humans were only ever weighed at birth" [p72]. However, some of the advantages he ascribes to streaming (for example, if a "stranger puts [your song] on a rap playlist that five more strangers follow may not turn out to be the start of anything bigger for either of you, but it *couldn't* before, and now it could" [p57]) could also be associated with downloading, and so aren't unique selling points.

His eye is not, however, uncritical: notably, he takes issue with the "myopically limited kind of surveillance that streaming services subject us to" which - like all recommendation engines which rely on "collaborative filtering" - looks at our past choices and the choices made by a larger group of other uses (which assumes that people who liked the same things in the past will share the same preferences in the future). He says that "they can only guess that we are exactly like the people they've seen before, but they never learned that much about *those* people either" [p41].

Given the vast scale of music available online, his main interest is the development of an automatic method for assigning a song or artist to a genre, hoping that'll aid each listener in finding more music that they like. To make that assignment more precise, he and his colleagues have been responsible for creating finer subdivisions between genres (i.e. the labels which used to be rock, pop, folk, classical and jazz, used for the "tersely labelled bins in record stores" [p192]) - at the most recent count, there are 5,000 of them. Perhaps this effort is akin to (in the words of Clive James) "drawing up plans for a centralised world laundry", but the result can be explored (and tested) in the authors everynoiseatonce website, and he's apparently used this in his exploration of more exotic musics - e.g. "post-WWI Czech tramping music, the wave of Yugoslavian mariachi bands in the 50s and 60s, and the thriving 80s reggae scene in Poland" [p176]. These selections might suggest that he's casting his net wider in order to make a point - thus, not to be too Anglo-centric, but there's hardly any mention in the book of music from the UK, for example.

Overall, the book is well-written, with a host of eye-catching detail - for example, his invention of "epicore" as the name for "bombastic music that sounds like it goes with portentous music trailers but doesn't" [p99], or pointing out that there's no term for a non-concept album, "which is how you know that non-conceptness is the Album Era's normativity" [p103], or recalling that "the pre-digital dark ages were so dark that radios didn't even have screens", meaning that if you heard a song you liked you had to wait and hope that "a human would stop playing music long enough to tell you, in words, what it was" in enough detail that you could go to the record shop "and hope they had it and it was on sale" [p106]. He snappily summarises the most important algorithms, saying "search jumps you into hyperspace, popularity organises the most likely places to land, and similarity lets you wander around looking at what's there" [p157].

Despite finding the book a stimulating read from a musical and technological viewpoint, I should add that, although I play music all the time, it's not streamed, but comes from a large (though finite) corpus on disk which gets added to every now and again. Some of the details in this interesting book have, however, prompted me to take another look at Spotify: perhaps I'll be a late adopter? Time will tell.