A group of more than 180 musical artists and human rights activists has sent an open letter to Spotify to ask the company to abandon a speech-powered song recommendation system. The company first described the system in a patent application in 2018, and received the official patent earlier this year.
The problem, according to the signatories, is that the system (as described) is both invasive and prone to racial and gender bias. As it stands, Spotify recommends new songs to users based on self-reported preferences and listening tendencies. The new system is designed to replace that “tedious” process with a more passive one that functions without any direct user input, analyzing ambient speech characteristics and background noise to queue up songs a user might want to listen to. For example, the system might play one song if the user is alone, and a different song if it thinks there are multiple people in the room. The system would also factor in traits like gender, mood, age, and accent when making recommendations.
The first of those features raises concerns about privacy, while the second raises concerns about the potential for various forms of bias. On the privacy front, the fact that the system relies on speech and background noise means that the system is always on, and that Spotify is always listening. The result is a massive corporate surveillance network that could give Spotify access to vast amounts of sensitive personal information, much of which would seemingly have little to do with song preferences. By the same token, the other people in the room may not be aware that they are being recorded (and therefore cannot give consent).
Meanwhile, any association between a trait like accent or gender and a certain type of music relies on (and perpetuates) biased and possibly discriminatory assumptions about what kinds of people listen to what kinds of music. Activists have previously noted that speech recognition algorithms display many of the same performance biases as facial recognition tech.
“Claiming to be able to infer someone’s taste in music based on their accent or detect their gender based on the sound of their voice is racist, transphobic, and just plain creepy,” said musician and human rights activist Evan Greer.
Spotify is yet to deploy the tool in any commercial capacity, and has stated that it has no plans to do so. The letter writers are nevertheless asking for a firmer commitment, and have demanded that the company promise to never use or monetize the system in any capacity.
For now, Spotify currently has around 356 million active users. Tom Morello, Talib Kweli, Laura Jane Grace, and Amnesty International are some of the more high-profile names attached to the open letter.