An Algorithm Could Be Your Next Favorite Artist
An Algorithm Could Be Your Next Favorite Artist
The Role of Streaming Services in Music Discovery
Machine learning has played a crucial role for Spotify and the music streaming industry more broadly. Before services like Spotify or iTunes, producers, artists and content distributors did not have the level of access to listener’s individual music preferences that they have now. Through listener’s real time data, streaming services can detect popular music trends and microtrends with much greater velocity and accuracy than previous music distribution channels, including radio or album sales, could ever accomplish. Streaming services’ unprecedented depth (individualized) and breadth (macrotrends) of data has provided fertile ground for emerging technologies such as machine learning to play a crucial role in content management.
As streaming services play a more prominent role in the music discovery and listening process, they compete and differentiate in their ability to deliver the right content to listeners. Streaming services therefore act as a marketplace where their goal is to adequately match music supply and demand. Streaming services that can accomplish this effectively, enter a virtuous cycle as a high number of quality artists (supply) attract more listeners (demand); then more listeners attract more artists to join the platform. This positive feedback loop incentivizes streaming services to invest heavily in tools such as machine learning. For instance, on October 2018, Apple acquired a machine learning startup called Asaii which lets producers and labels “discover, track and manage talent” by analyzing listener’s data from streaming services and social media. [1] The company claims that, through their machine learning algorithms, they are able find artists 10 weeks before chart.
Machine Learning as a Competitive Differentiator
Management at Spotify recognized early on that optimizing its marketplace capabilities would be a key source of differentiation among streaming services as customers would derive more value from the platform. In 2013, they acquired a startup called Echo Nest which uses machine learning and acoustic analysis to create tailored recommendation to users based on their individual listening. [4] Recommendations come in the form of notifications and bespoke curated playlists such as “New Music Friday” and “Discover Weekly.” The relevance of the content attracts users to the platform which stresses the need to have accurate and ever improving machine learning technologies. More recently, Spotify has continued improving on its machine learning algorithms and in August 2018, it filed a patent on technology that identifies personality traits based on user’s listening history. The system assigns a “first personality trait” which then provides personalized content on this trait. [2]
While tailored content might still prove to be a key differentiator in the near term, competitors, like Apple through its acquisition of Asaii for instance, are developing similar capabilities pressuring Spotify to find other ways to use machine learning to differentiate its product offering. In order to do this, Spotify hired researcher Francios Pachet to develop tools that will help artists in the creative process. [3] Dr. Pachet has focused his research on how machine learning can help computers understand musical style and composition. More importantly, machine learning has started to be recognized more broadly as a viable way to augment the artist’s creative process. In the Journal of New Music Research, researchers Sturm et al partnered with musicians to utilize existing machine learning models such as folk-rnn, DeepBach, among others, to complement their creative process. While the machine generated music did not completely remove the artist’s creative process, it did help build on the creative process. One artist noted that, “[content] was ‘foreign’ to me, but served as… inspiration as well as a way of developing my own compositional language.”[3] Through Spotify, Dr. Pachet can access data on listener’s preferences to refine processes and identify what elements can be appealing to the targeted listener. Competitively speaking, Spotify would be providing tools to artists so they can create more meaningful content for listener’s and increase the overall value of the marketplace.
folk-rnn machine learning content; OBT modified content by the artist based on machine learning
Considerations of Historical Data on Innovation
While there seems to be increasing evidence that machine learning can provide benefits to all stakeholders, the output from machine learning relies on finite historical data, which might adversely impact the diversity of the content generated. Using machine learning to identify the essence of ‘hits’, perpetuates those key components potentially leading to repetitive or stale content. If musicians are planning to use machine learning to develop content, it is important to maintain a human element such that new and innovative content is not lost in the process.
The Beginning of the End for Emotionally Inspired Music?
Ultimately, the next wave of machine learning technology seems to have promising benefits for Spotify, but are listener’s better off with content that was engineered by a machine rather than hand-crafted out of the inspiration of a musician? (word count 798)
Endnotes
[1] Boland, Hannah. 2018. The Telegraph. October 15. Accessed November 10, 2018. https://www.telegraph.co.uk/technology/2018/10/15/apple-buys-music-analytics-firmasaii-help-spot-emerging-talent/.
[2] Gibson, Clay, Will Shapiro, Santiago Gill, Ian Anderson, Margreth Mpossi, Oguz Semerci, and Scott Wolf. 2018. Methods and Systems for Personalizing User Experience Based on Personality Traits . Stockholm, Sweden Patent 20180248978. April 9.
[3] Sturm, Bob L., Oded Ben-Tal, Úna Monaghan, Nick Collins, Dorien Herremans, Elaine Chew, Gaëtan Hadjeres, Emmanuel Deruty, and François Pachet. 2018. “Machine Learning Research That Matters for Music Creation: A Case Study.” Journal of a New Music Research 1-20.
[4] Titlow, John Paul. 2017. FastCompany. July 13. Accessed November 10, 2018. https://www.fastcompany.com/40439000/why-did-spotify-hire-this-expert-in-music-making-ai.
Such an important use of machine learning given it touches most of us everyday, thanks for raising it. I totally agree with how you describe the circle of benefits as more consumers listen, then more artists are inclined to join the site. It was really eye-opening to read about a future whereby songs are written based off popular beats. As consumers we often don’t notice the similarity in beats, but it is really interesting to know that music is so simmilar. I had not realised machine learning could help impact this. You raise very important questions for the future of music- should they be written based off an algorithm or use human creativity? I totally agree with your consensus that human input is still valuable because the essence of music itself is that its creative.
As a daily Spotify user, I found this super interesting (and well-written)! It was only just recently that I realized Spotify was using machine learning to recommend these “Daily Mixes” that suited me perfectly. It’s so spot-on that I rarely create my own playlists now. However, I do recognize that I tend to listen to the same genre of music because of that and rarely venture into other genres. Sometimes I like that, other times I’m unsure. So I guess this goes back to the concern you raised – is this putting a box around my musical interests? Also, although Spotify is using the same concept and technology and just applying to the musician side (vs. the consumer side), somehow I feel less open to the idea of musicians/artists/composers using technology to create music based on listener preferences. Perhaps I’m idealistic in thinking that music should be about personal expression and when you use a machine or technology to create it, is it still personal? Who is the actual creator – the machine or the person using the machine? Also, there have been many cases of song or tune plagiarism due to similar melodies or other aspects. How does the technology protect users from falling into these issues? Just some thoughts sparked from your essay!