Facebook will pay users up to $5 for voice recordings via its Viewpoints app to improve its speech recognition tech; recordings won't be tied to user profiles
But it won't pay much — Facebook will offer to pay some users for voice recordings that will be used to help improve …
Context & Ripple Effects
Facebook's Viewpoints app, launched as a paid market-research platform in November, is now being used to source training data: users can earn up to $5 for voice recordings that feed the company's speech recognition work. The move lands five months after Bloomberg reported Facebook had paid hundreds of contractors to transcribe user audio without clear user awareness — work the company said it stopped.
The framing is deliberate: recordings collected through Viewpoints won't be tied to user profiles, converting what was a passive-collection liability into an explicit, compensated transaction. It also serves the voice roadmap Facebook has pursued since its Aloha speech-recognition project and speaker patents, and the voice dictation and calling features Messenger has been testing.
First-order effects
- US Viewpoints users gain a new paid task — voice recordings for up to $5 — making them compensated suppliers of speech-training data rather than unknowing sources of it.
- Facebook's speech recognition teams get a consent-based data pipeline to replace the contractor transcription program the company halted after the Bloomberg report.
Second-order effects
- The 'not tied to profiles' commitment becomes a competitive talking point: Facebook can pitch voice features like Messenger dictation and its speaker ambitions as built on opted-in data, pressuring rivals' passive collection practices by contrast.
- Paying users establishes a price anchor for voice data — small, but a precedent that data collection is a transaction, which other platforms sourcing assistant training data must now either match or justify avoiding.
Third-order effects
- If the pattern holds, AI training data shifts from covert extraction toward explicit paid-consent markets, with apps like Viewpoints functioning as procurement channels and regulators treating undisclosed collection as the outlier rather than the norm.
- Voice assistants' viability increasingly depends on this consent architecture: the platforms that can lawfully and cheaply source labeled audio — through paid user programs rather than contractor leaks — hold the training-data advantage.
The trend: Consumer voice data is moving from passively collected to explicitly purchased, with paid research apps becoming procurement channels for speech-recognition training sets.