Facebook Will Pay You (A Little) to Record Your Voice
Facebook will pay for people’s voice recordings to help train the speech recognition engine for its Portal smart display. The tech giant is soliciting the audio through its Viewpoints market research app, launched late last year.
The Sound of Money
The Viewpoints app solicits information through surveys and beta tests. For the voice recordings, Facebook asks participants to speak the names of several friends and say, “Hey Portal,” the wake word for Facebook’s smart display. Each time a user does a set of recordings, they earn points on Viewpoints. Five recordings are worth enough points to get $5 cash. Every recording contributes to improving how well the Portal AI recognizes its wake word and people’s names. Considering how much Facebook markets Portal as a video phone, those are critical parts of speech for it to understand.
It’s not a lot of money, but it’s notable any time a tech company that profits from user data offers to pay for it. There’s also the question of scale. Machine learning relies on AI processing as much data as possible. The Viewpoints recordings may garner only a tiny percentage of Facebook’s userbase, but even a small fraction of Facebook’s billions of users would be useful for Portal’s improvement. The economics of the new offer is similar to Fiverr or Amazon’s Mechanical Turk, marketplaces for outsourcing a whole range of menial tasks, including voice recordings. People might sign up for a project hosted on either platform, although Fiverr also has a long list of people offering their voice recording services at prices as low as on Viewpoints.
Privacy and Value
Every voice assistant developer needs people’s voices to improve how well the AI understands and interacts with users. But, when that audio is gathered less directly than Facebook’s offer, many people become anxious about privacy. Reports about how Amazon collected and shared Alexa recordings with contractors as part of its quality control and improvement program kicked off a cascade of similar revelations for months regarding Google Assistant, then Apple’s Siri, Microsoft’s Cortana, and more. Facebook faced similar controversy of its own when stories of how it hired contractors to transcribe audio sent through its Messenger app.
Most voice assistant developers paused their programs, revising how they operate. The changes increased transparency, but the underlying issue of people not always knowing what happens to their audio recordings is still there.
Facebook’s new offer circumvents that by explicitly requesting the recording and paying for it. People who participate know precisely what they are sending and are compensated by Facebook. According to Viewpoints’ terms and conditions, any data collected may be shared with research partners, but users will be told about it if that happens. Privacy continues to be the biggest single reason people say they are reluctant to use voice assistants. Facebook and other developers will need to keep working on ways to get the data they need to build and improve their AI platforms without scaring away potential users.