Since the advent of streaming radio, streaming music companies have been vying to provide the best user experience while leveraging the latest in modern technologies. Some online music companies went the route of giving the user total control over what they listened to, while others, such as Pandora, used listener recommendations to build user-influenced but ultimately computer-generated radio stations.

Now, the two methods of music delivery might be merging as companies like Pandora and Spotify work to expand their user bases.

How Pandora Uses Machine Learning

The original Pandora music delivery model uses artificial intelligence to capture “thumbs up” data from users to match listeners with songs they might like. The data is enhanced by the human-curated Music Genome Project, a database of up to 50 million songs that tags music based on more than 450 attributes.

Pandora’s original interface did not give listeners 100% control over what they listened to. You could not, for example, select an album by a particular artist and listen to it all the way through. Instead, you might select the artist (such as “The Beatles”), create a “radio station” called “Beatles Radio” with the Beatles as the main “seed” for the station, and then Pandora’s AI would select songs for your station based on various algorithms.

While listening to the station, you might “thumb up” a song or skip it altogether (thus removing it from your station permanently). Each like and dislike would further personalize and tailor the station to your tastes. Pandora’s machine learning would ultimately keep winnowing down your music until you had a station that truly reflected what you wanted to hear.

How It’s Different from the Direct Music Delivery Approach

Pandora’s competitor, Spotify, originally focused on a simpler model: Just give users exactly what they ask for. With a huge database of albums and songs, subscribers to Spotify could just pick an album or a song and play exactly that album or song. Users could of course generate playlists and even share them with others. (President Obama actually created a “Summer Playlist” for Spotify while in office.)

Spotify’s direct approach would give a user exactly what they wanted, but would not necessarily introduce listeners to new songs and artists the way Pandora’s machine learning might. Spotify has used “collaborate filtering” that uses AI to build associations based on user choices, but that music isn’t inserted into an existing Spotify playlist without user permission.

Record companies, of course, want users to try out new music, and they work with music streaming companies to encourage recommendations and cross-promotions, even without machine learning tools. But for a new artist, completely hand-picked music doesn’t easily open doors to new fans.

Getting the Best of Both Musical Worlds

With a lot of competition in the music streaming market, both Pandora and Spotify are expanding their business models to include both artificial intelligence and user control.

In March of 2017, Pandora launched Pandora Premium, which includes many features that Spotify has had, including the ability to play entire albums and create 100% user-controlled playlists. The new features are in addition to Pandora’s existing machine learning radio station generator.

While this may seem like a positive trend for users, it comes in response to Pandora’s financial troubles.

Spotify, on the other hand, is expanding to create more Pandora-like features. In May of 2017, it announced the acquisition of a machine learning startup called Niland. Using advanced machine learning technology, each song will be broken down into multiple components, including the instruments used in the track.

Spotify’s machine learning system won’t include the human element of Pandora’s Music Genome Project, and whether the AI can provide truly pertinent music recommendations remains to be seen. Spotify is also losing money, in part due to the high cost of licensing fees, which eat about 85% of its revenue.

Perhaps, in the future, music streaming users will be able to do their own rudimentary AI “programming” to tell the computer how, exactly, they would like their recommendations generated.

Does Curated Music Still Have a Place?

With all the focus on user choice and granular recommendations based off of computerized algorithms, are music deejays permanently out of jobs?

Despite the trend for artificial intelligence and machine learning, some music services still offer human-curated – or at least, curated – playlists.

Music Choice, which provides music channels to cable television customers directly through televisions, as well as via the web and mobile apps, boasts that their stations have human-selected music. Users can’t control what they listen to, other than choosing the channel.

Calm Radio is an online music streaming service that only offers “curated” channels, although they don’t advertise them as being human-selected. Currently, the service does not offer any way to “thumb up” a liked song, although songs can be skipped (which does not affect whether they will play again in the future).

Calm Radio’s selling point is that it offers music as well as “atmosphere” channels that are geared towards reducing stress and providing a “calm” listening experience. (The service does offer some pop music channels, as well.) The atmosphere channels include nature sounds like thunderstorms, bird song, and ocean waves, as well as white noise and other sounds that people find soothing, including, strangely enough, the sound of a city sweeper.

The technology that sets Calm Radio apart is the ability (depending on the interface used) to layer channels on top of one another. In this way, a user can play a classical station with a rainstorm in the background.

The Future of Online Music Streaming

Whether or not companies like Pandora and Spotify survive in the long-term, the future of music streaming is sure to include user recommendations and AI-generated content. As the technology becomes more and more sophisticated, machine learning algorithms will soon be able to predict future music choices of listeners as they age and their tastes evolve.

Image Source: Adobe Stock