Discovering classical music

How to revitalize the classical music ecosystem

David P. Anderson
1 January 2024

Summary: Current and future classical music platforms (like YouTube, Musescore, IMSLP, and Groupmuse) should offer better tools for discovering music and musicians, including tools that infer and reflect individual musical taste. This would: Help works by living composers get played, recorded, and performed. Help listeners and performers find music (possibly obscure) that resonates with their individual taste. Help amateur performers get heard. Classical music would then become a more vibrant and dynamic ecosystem. Introduction People interact with classical music* in many interrelated ways. They listen to recorded music at home and live music at concerts; they play and perform music; they compose music; they discuss music with other people. For some people, classical music is a pleasant sonic background; for others it's a source of strong emotional and intellectual response. This essay is concerned with the latter group. For these people, musical activities can provide various psychological benefits, which I call 'rewards'. Sub-essay: Rewards in classical music I'm an amateur classical pianist. I live in the SF Bay Area, which has a rich musical culture. There's a conservatory, a symphony and opera company, and many concerts, big and small; I go to lots of them. I belong to an amateur music group where I perform every month or two. My social life is centered largely around music. So my musical life is pretty good. But it could be a lot better: I'd like to play more chamber music, but it's hard to find collaborators. I'd like to discover new music, perhaps by living composers, to play and listen to. I'd like to hear more live music - especially modern music. All these things exist - the problem is how to discover them. Classical music needs better discovery mechanisms. I know great musicians who have fairly sparse musical lives. They're undiscovered, and the world is a poorer place for it. Internet-based websites and smartphone apps (which I collectively call 'platforms') have greatly enhanced the world of classical music. IMSLP and YouTube make millions of scores and recordings available online; Groupmuse has enabled a new house-concert culture; Meetup has enabled social activities like my amateur music group. This essay proposes ways to enhance these platforms (and create new ones) to further improve our music lives: to provide more rewards, and to strengthen the classical music ecosystem. The central ideas are: Music platforms should offer better tools for discovery (of compositions, recordings, concerts, and people), including tools that reflect individual musical taste. Music platforms should form a consortium, the Classical Music Index, to standardize classical music metadata, perhaps using the current IMSLP metadata as a starting point. Music platforms should form a consortium, the Music Preference Service, to share ratings and usage data, to infer individual musical taste, and to make recommendations. This is shown below. Proposed new components are in heavy boxes. Concert Finder and Music Match are described below. Note: this essay is concerned with what could and should exist. How to get there from here is a separate issue. Persuading competing companies to cooperate is a challenge. The essay covers a lot of ground. To make it more manageable I've moved some of it into sub-essays. I hope you have the patience to read (or at least skim) it all. Please contact me if you have feedback. Classes of musicians It's useful to classify musicians (performers and composers) in terms of finances: Class A: musicians who make a comfortable living from performing and/or composing. They have full-time orchestra gigs, or contracts with record companies, music publishers, talent agencies etc. Class B: musicians struggling to make a living from performing and/or composing. They play weddings and company parties, and have no long-term contracts. Class C: musicians with a musical day job (generally teaching) who perform or compose on the side. Class D: musicians with a non-musical day job (around here, generally a high-paying tech job), and who would like to perform publicly or to have their compositions performed. This classification is not a ranking by talent or accomplishment. I know class D pianists who play better (to my ear) than many class A pianists. But few people ever hear them play. I know class D composers who are amazing, but their music never gets performed. The sizes of these groups vary widely. In the U.S. there are perhaps a thousand in class A, a few thousand in class B, tens of thousands in class C, and perhaps a million in class D. I'm especially interested in class C and D musicians; they're a huge pool of talent, and are underserved by the current ecosystem. Musicians in class A and B care a lot about getting paid. Class C musicians care, but less so. Class D musicians don't care about getting paid; in some cases (amateur choirs and orchestras) they pay for the chance to perform. Money and rewards are largely orthogonal. The biggest rewards are often free. Everyone wants rewards. If you're trying to make a living from music, you want money too, maybe more than you want rewards. It would be nice if more people could make a living from music. But that's outside the scope of this essay; I focus on rewards, not money. Musical ecosystems Each musical genre has its own 'ecosystem' of audience, performers, and composers. These ecosystems differ widely. Consider rock music. People listen to a mixture of old and new rock. New songs and bands are constantly arising - garage bands and local bar bands. There's always exciting new music; you can find it on YouTube, Bandcamp, and so on. We listen to it, but we also listen to rock from past decades - maybe 50/50. The best of the new stuff sticks around, and we'll be listening to it decades from now. This seems like a healthy ecosystem. There's a sense of lineage and history; musicians build on what previous musicians have done. But it also moves forward and evolves, driven by grass-roots activity. Today's classical music ecosystem lacks these properties. Instead: It's elitist: it revolves around big stars, big venues, and big money. Self-proclaimed 'experts' tell us what we should like. There's hero worship: unless you're Yuja Wang or the Kronos Quartet, why even bother? Virtuosity is emphasized over expression. We hear the same pieces - the 'standard repertoire' - over and over. Obscure or new music doesn't get played much. It's mostly non-social. People go to concerts and go home, not interacting with the performers or other audience members. But I can imagine a (non-elitist) classical music ecosystem in which There are lots of concerts, many of them in houses, ranging in size down to a few people in a living room. Some charge, some are free. Lots of these concerts feature Class C and D performers. Concerts (of all sizes) commonly feature new music by class C and D composers. Good composers get heard without having to promote themselves. Concerts (especially the small ones) are social as well as musical events. Performers can easily find other performers to play with. Composers can easily find performers to play their music, and conversely. There would be far more music and musicians than today. It would be reminiscent of the 1800s, before radio and recordings, when lots of houses had pianos, and live music was a staple of family and social life. People tend to divide classical music into 'old' and 'new', and to think of new classical music as abstract, emotionally arid, and unrewarding. They think all the great classical music was written one or two centuries ago. I disagree. I think there are (or could be) contemporary composers at the level of Bach, Beethoven, and Chopin. We don't know about them because it's hard to discover their music. Online search in general Many popular Internet platforms provide access to a set of "items". Amazon and eBay let you find consumer products. YouTube and Netflix let you find videos. Companies and professionals find each other on LinkedIn; single people find each other on OKCupid. In each case, items have various attributes: name, description, category, size, price, etc. Platforms offer 'search features' to let users find items. These features can be roughly categorized as: Textual search: usually string matching on item name. This is useful if you go to the site knowing what item you're looking for - perhaps a particular product. Attribute search: the user can explore a hierarchy of item attributes (e.g. apparel/mens/outerware) or do a SQL-type query on multiple attributes. Attribute linkage: When the platform shows an item, it shows a list of items with similar attributes (a streamlined form of attribute search). Social linkage: Platforms keep track of user interactions with items: what they viewed, what they bought. They may also let users review and rate items. This data can be used to facilitate discovery: When the platform shows an item, it can show a list of items linked by user interactions. For example: "People who bought this item also bought these items". Or it might show a list of users who bought or reviewed the item, and you can click to see what else they like. History linkage: the platform shows items that are linked (e.g. by attributes) to items the user has expressed an interest in (e.g. viewed or bought). Collaborative filtering. The platform uses interaction data (e.g. item ratings over the whole population of users) to predict what items a particular user will like. This is related to social and history linkage but is more powerful; it can can yield items with no direct link to the user, and (ideally) which match the user's individual taste. These mechanisms can be combined; for example, the results of an attribute search could be ordered by social linkage (show highly-rated items first) or by collaborative filtering. History linkage and collaborative filtering are 'personalized': what a user is shown depends on that user, ideally reflecting their tastes or interests. For some things - say, kitchen appliances - personalization doesn't matter much. If a coffee grinder is good, it's good for everyone. But for classical music, where individual taste is diverse, personalization matters a great deal. Terminology I use search to refer to the whole spectrum between Lookup: where the user arrives at the platform knowing exactly what item they want. Discovery: where the user uses the platform to find items they don't currently know about. Suppose there's a large space X of items, and a user will like (because of their individual taste) only a small subset Y ⊂ X. A discovery mechanism is called Efficient if it mostly shows items that are in Y; otherwise the user has to wade through a lot of items before finding something they like. Complete if it's able to show all the items in Y. Our challenge is to find music discovery mechanisms that are both efficient and complete. Search in classical music Classical music involves several types of items: Compositions Scores Recordings People (composers and/or performers) Ensembles Concert venues Concerts, past and future The notions of search and discovery apply to all these types. A performer might look for a composition to play, a (living) composer to commission, a choir to join, or a concert to attend. A concert venue might search for performers or ensembles. And so on. Items can have various attributes: Locations of concerts and people. Dates (birth/death of composers, dates of composition, etc.). Genres Category (Sonata, Mass, Symphony, etc.) Instrumentation Items are connected in various ways: a recording is linked to a composition, a composition has a composer, and so on. An item's attributes, and the items it's linked to, are called its "metadata". The search mechanisms described in the previous section above all apply to music. For example, 'attribute linkage' would mean that a web site's page for a Chopin piano piece could have links to other piano pieces by Chopin, piano pieces by other 19th-century French composers, and so on. Classical music has several important discovery-related properties: Size. There's a huge amount of classical music - millions of compositions, more than you can hear in a lifetime. Scattering of items. Items are not all in one place. For example, lots of pieces by obscure or unknown composers are unpublished; they may exist only in the composer's notebook or computer. Individual taste. A person may love only a tiny fraction of the millions of classical pieces. This set may be different for everyone. It will generally include obscure or unknown pieces. Change. Most musicians don't want to hear and play the same pieces repeatedly; they want new things occasionally. For example, I'm always looking for new piano pieces to play. I have fairly unusual and narrow musical taste. My piano technique has strengths and limitations. I'm sure that somewhere out there - maybe on the laptop of an unknown composer in Kazakhstan - there's a piece that's perfect for me; Learning it and playing it for people would give me a big reward; it would also give the composer a reward. How can I discover this piece? Before the Internet, there were various ways for me to discover compositions: I hear a new piece at a concert or on the radio. I find an interesting-looking score while browsing at a (physical) sheet music store. A friend tells me about a piece or composer, or I find one while browsing a friend's CD collection. None of these lets me find that piece in Kazakhstan - they're incomplete. They're also generally inefficient. The situation is similar for other item types: If there are 1000 violinists in the Bay Area, how can I find the ones I'll like accompanying? If there are 100 concerts a night, how can I find the 1 or 2 that I'll like? Conversely, if I'm a class B/C/D performer, how can I find the people who will like to hear me play? And so on. The Internet creates new possible avenues for music discovery. There are lots of platforms related to classical music: IMSLP, YouTube, Spotify, Groupmuse, and so on. They provide easy access to lots of musical information. They all provide lookup features. Some provide discovery tools, but most do not. These platforms expose a much larger range of music than was available before. IMSLP and MuseScore allow the unknown Kazakh composer to make his work available to the world without a publisher. But I still need a way to discover it efficiently. So: how can we improve the processes of musical discovery? First, platforms should provide discovery tools that reflect individual taste in music, using collaborative filtering based on user ratings and interaction data. The model of a person's musical taste should be based on data from multiple platforms. The pieces suggested by IMSLP should reflect not only my interactions on IMSLP, but also what I listened to (and maybe rated) on YouTube and other platforms. Sub-essay: Musical taste and personalization So platforms should pool their personalization data. For this to be possible, they must agree on a standard way of identifying items (like compositions and composers). They must form a consortium (which I call the "Classical Music Index") for this purpose. Sub-essay: Standardizing classical music metadata Finally, platforms should provide a variety of social features, supporting communication (group, individual, public, private) between musicians. This is needed for discovery mechanisms involving social linkage. Sub-essay: Socializing classical music platforms How to improve music platforms There are lots of music platforms, differing in several ways: They focus on different item types: scores, recordings, or concerts. Some focus on Class A musicians, and get their content from music publishers or record companies. Others allow all musician (including Class D) to upload content. I discuss these classes of platforms in sub-essays; see below. But in general they can all be greatly improved - in particular, none of them provide good discovery features and few provide social features. They should all do the following: Where the platform shows an item (for example a composition or recording) it should let you rate it and/or comment on it. Where the platform shows an item it should show you a summary of its ratings and comments. Where the platform shows a list of items (say, the results of a search) if should order the results by decreasing personalized predicted rating (if possible) or by decreasing average rating. Where the platform shows an item it should show a list of related items: those with correlated ratings (if available) or with correlated access. Where the platform shows an item it should include a list of people who like it or commented on it. These should link to user profiles, showing what else they like. Each of these features provides an avenue for discovery. I discuss the various platform types in sub-essays: Sub-essay: Discovering compositions Sub-essay: Discovering recordings Sub-essay: Discovering concerts Sub-essay: Discovering people These sub-essays discuss how existing platforms could be improved; they also propose music platforms that don't currently exist. * By 'classical music' I mean notated music with lots of subtlety, complexity, and/or expression, including old music (Bach, Beethoven, Brahms), 20th-century music, and current and future contemporary music. However, the ideas in this essay apply (to varying degrees) to all kinds of music.