The major tech companies making voice assistants all faced consumer backlash in the past several months over privacy issues. While many took steps to give users more control over participating in human review programs, the exposés also revealed that many smart speakers record their users at times when the individual hadn’t uttered a wake word. New research from a group of universities explored the extent to which popular TV shows can accidentally turn on a smart speaker’s microphone.
The researchers from Northeastern University and Imperial College London played 125 hours of Netflix content from twelve different programs, containing voices from a range of demographic groups and talking about a variety of topics. During the experiments, the speakers tested activated between 1.5 and 19 times per 24 hours.
Apple’s HomePod and the Harman Kardon Invoke from Microsoft had the highest rates of activations. The more dialogue-heavy TV shows Gilmore Girls and The Office created the majority of accidental activations. Each of the dozen programs in the test sparked at least one unintentional activation.
The ongoing research shows how much work remains to be done in finessing AI understanding of wake words and the ability to distinguish between similar-sounding phrases.