Spotify’s Speech Recognition Patent: Should We Be Worried?
Spotify has patented tech that it claims can recognize your emotions and characteristics via speech recognition. Will they use it? Is it even legal?
On January 12, Spotify had a patent approved. The patent would supposedly help its users receive personalised playlists without the need to "tediously input answers to multiple queries" about their preferences and characteristics.
On Friday, digital rights group Access Now sent Spotify's CEO a letter, calling the patented product a "dangerous" "violation of privacy and other human rights."
What's the patent for?
Spotify proposes a technology that would use speech recognition to "retrieve content metadata" about people's emotional states, gender, and age, and use this information to make music recommendations.
The idea has caused considerable alarm, not least among digital rights advocates at Access Now, who are urging Spotify not to use the technology.
What's the problem with it?
Access Now has four main issues with Spotify's speech-spying specification:
- Monitoring people's emotions puts Spotify in a "dangerous position of power" over them. The app would have an incentive to reinforce people's emotional states—to keep depressed people depressed.
- Attempting to infer a person's gender through their speech patterns could lead to discrimination and the reinforcement of gender stereotyping.
- If, as has been reported, the device is "always on", this would violate people's privacy.
- There is a huge security risk involved in recording people's conversations.
That does sound quite bad.
There are two important things to bear in mind here.
First, Spotify might not go along with this. Companies sometimes patent ideas speculatively, or to prevent their competitors from doing so.
Google once patented a tattoo that would embed a microphone into your neck. It wouldn't actually bring that to market, right? Right...?
Second, this tech might not actually be very good. Emotion recognition is often said to be overhyped by AI people. If it doesn't work, Spotify presumably won't use it.
Would this even be legal?
In the U.S., I think Spotify could get away with this, but it would have to avoid creating biometric "voiceprints" without consent. Doing so would violate Illinois' Biometric Information Processing Act.
Under the GDPR, there are rules around taking decisions based on "automated processing", but only where such decisions have "legal or similarly significant effect".
If Spotify detected you were feeling happy but it decided to play Radiohead's How to Disappear Completely on repeat, this would have a "significant" effect on your mood—but it probably wouldn't meet the above threshold.
However, interpreting someone's speech to draw inferences about them is certainly "profiling", and there are restrictions on profiling under the GDPR.
I think Spotify's idea would be likely to fail an honestly-conducted GDPR data protection impact assessment, particularly if the feature is "always on", and particularly if it is on by default.
On this basis, I would bet that Spotify's creepy feature won't hit the market for quite a few years—in Europe, at least.