Before long, it won’t be enough for music services to play songs you might enjoy. Instead, the music platforms of the future will know exactly what you want to hear based on your mood. But how?
The latest foray into mood-powered music comes from Rok Mobile. When the music-focused mobile carrier launches in a few weeks, it will be powered by Gracenote’s Rhythm platform. This will give users of the Rok music service the ability to use a mood grid to play songs based on how they’re feeling. The company’s Rhythm platform has been categorizing song moods for a while, but this is the first time it’s being used publicly.
Rok Mobile isn’t the first company to tackle mood-based music curation. Beats famously introduced its Sentence feature, which lets you dictate a playlist by playing a Mad Libs-like fill-in-the-blank. Songza also built its service on playing music depending on what you’re doing right now. Moodagent is one of the more visual music apps, using sliders to adjust the amount of “angry” or “happy.”
Rok Mobile will also be visual in its approach. Users of the mobile service’s app will see different squares defined by extremes on the top, bottom, left, and right. The X axis is defined by intensity ranging from calm to energetic, while the Y axis is emotion ranging from dark to positive.
“Music evokes memories and emotions, it helps define how you are feeling at this very moment,“ says Ty Roberts, chief strategy officer and cofounder of Gracenote. ”By assigning specific moods to every song, Gracenote is finding commonalities that cross artists, genres, and eras. This allows us to bring together songs that share similarities to match what you want to listen to right now.”
When Gracenote gets new music, there’s a few different ways that it initially breaks it down and gets it into its system. If it’s a new song from a new artist then the human editorial team–or “musicologists” as they’re called–listens and assigns a basic artist profile.
A new artist with new music: First, a new artist with a new song gets tagged with a genre profile. If a song spans more than one genre, it’s weighted, such as 80% contemporary R&B and 20% hip-hop. Next, a song is tagged with its origin. For example, The Killers are from–and associated with–Las Vegas.
An era is then assigned. Because we’re talking about new artists with new songs, this is typically the current era (i.e., early 2010’s). Then there’s artist-type information, which includes gender and/or composition of the group.
Existing artist with new music: If an existing artist produces a new song, there should already be an artist profile and it’s typically carried over to the new song.
Complex artists: For the few artists that truly do change from one release to the next, a new song is listened to and assigned another genre from the broader artist profile. If the genre isn’t already in the artist’s profile it’s given a new genre outside that profile.
Once songs are initially categorized based on the process described above, they are then ready to be dissected for insights about their mood.
Gracenote’s feature-extraction system analyzes the sound waves of an audio recording for features that mathematically describe qualities like harmony, rhythm, timbre, and melody. The system then takes those audio features and puts them into its sonic mood classifier algorithm, which looks for patterns that can help the system classify tracks by mood.
A song is weighted against a mood vector containing 101 different moods. Based on the weighting, each song is then assigned a spot on the mood grid.
Those 101 public-facing moods are getting distilled further into 25 different moods, in the case of Rok Mobile. Examples of the different types of moods include defining “peaceful” as pastoral/serene, delicate/tranquil, reverent/healing, and quiet/introspective.
Having 300 different moods internally used moods helps to train the algorithm and improve accuracy. By having so many different options, the goal is consistency.
It doesn’t really matter if someone gets the 100% exact mood they’re looking for, as long as it’s close. The real issue is having all the songs that play being closely aligned. The more categories there are, the more likely that similar songs get grouped together rather than forcing a wider range of songs into few groups.