What do wake words do?

Tags
experiments
Contributor
James Parker
Date
March 1, 2024
Folgezettel
3c

The term ‘wake word’ was popularised by Amazon with the launch of its Echo smart speaker in 2014. Alexa wasn’t just the name it gave to its new digital voice assistant. It was the word you used to ‘wake’ it. In Amazon’s framing, the ‘wake word’ was a technical solution to a practical problem. For the Echo to be ‘voice first’, it needed a way of distinguishing so-called ‘general speech’ from ‘commands intended to be executed by the device’.

But the wake word wasn’t primarily a technical innovation. It was a trojan horse, designed to inveigle listening devices into our homes via a fantasy of consent, barely eighteen months after the Snowden revelations. And the moment the wake word was introduced, it started to be eroded. Saying Alexa’s name every time you want to interact with it wasn’t ‘natural’. So Amazon introduced ‘conversation mode’, which leaves Alexa ‘on’ as long as you’re in view of your device’s camera and looking toward the screen.’ In a pandemic, washing your hands became unusually important. So a 2020 update to the Apple Watch used ‘machine-learning models to determine motion, which appears to be hand-washing, and then use audio to confirm the sound of running water or squishing soap in your hands.’ All this with a view to helping you ‘keep going for the amount of time recommended by global health organisations.’ Here is a future in which the wake word is no longer something you say but something you do. Your body will consent on your behalf.

This movement towards wakewordlessness is accompanied by a simultaneous tendency towards wake word proliferation. Now, every company wants a voice assistant, and a ‘customized, branded wake word’ to go with it. Why? To ‘deepen user engagement, increase brand affinity, and inspire loyalty when your users ask for you by name every time they interact with your product.’ (Soundhound)

The wake word is so much more interesting and important than people give it credit for. It gives rise to so many questions. If a wake word isn’t exactly a technology, for instance, what is it? How does the wake word work? And where did it come from? What does the wake word wake? What kinds of legal, political and economic relations does it authorise? How is it teaching us to be and to speak? And what happens when a wake word detection system misfires? Or when we do? What kinds of accident does the wake word invent?

Experiment

Engage with some or all of the materials above. Think about your previous experiences with wake words. Try them out again: in your home, on the street, on your phone, in a car. Practice using a voice assistant the same way you would practice singing or playing guitar. Make yourself a virtuoso. Make mistakes. Take notes. Read the terms and conditions. Report back to the group on your experience, and how it related to the materials you engaged with.

Once you have have spent time thinking and talking with each other, gather around a collection of wake word triggered devices. Maybe three or four. Start invoking them. Ok Google. Hey Siri. Alexa. Computer. Cortana. Bixby. Try to keep the channels open. Keep the devices awake and listening. Keep pouring data in. Donate the sounds of your voices and those of the room you are in.

Take a moment to reflect on the experience together.

Resources

Wilpon et al (1989), ‘Application of hidden Markov models for recognition of a limited set of words in unconstrained speech’, International Conference on Acoustics, Speech, and Signal Processing

Salvador et al (for Rawles LLC, on behalf of Amazon) (2012), ‘Wake word Evaluation’, US Patent

Branden Hookway (2014) Interface (MIT Press)

Lauren Lee McCarthy (2017) Lauren

Mark Andrejevic (2018) Operational Listening

Machine Listening (2022) After Words