Amazon to record you before you say “Alexa”!

Amazon to record you before you say “Alexa”!

Pre-Wakeword Speech Processing

Speech recognition has evolved to a point that we can now talk to our devices. All you need is to speak aloud the wake word for your device, and dish out the command. From playing music to making calls, and setting reminders, we can now do it all with just our voices.

But, you need to first say the wake-word aloud to make your device start listening. You cannot just talk to your device as you would to another person. The command needs to be in proper syntax. Amazon is looking to address this very problem in a recently published patent.

Per the patent, Alexa devices, like Echo speakers will listen and record your speech even before you say Alexa. Your devices will capture your speech before and after you say the wake-word to determine the complete command. To do so, your speech is buffered and markers are inserted to indicate phrases and sentences. Then, when you say “Alexa”, Amazon will analyze the complete phrase containing the wake-word using automatic speech recognition (ASR) and natural language understanding (NLU). So, you will be able to give commands like “Play music for me Alexa from Travis Scott’s Astroworld”.

But this potentially increases privacy concerns. Although your Echo speaker still actively listens to your voice, it is only to determine if you’ve said “Alexa”. Your speech is not constantly recorded and analyzed. But as Amazon says in the patent, Amazon “does not necessarily perform full ASR processing on all detected speech, thus addressing privacy concerns associated with an “always on” speech processing system”. What this means is that not all speech is recorded and analyzed.

This technology looks like the way forward in our interactions with our devices. But the potential privacy concerns cannot be overlooked. We just hope that all the captured and recorded data is stored extra securely.

Patent Information
Publication Number: US20190156818A1
Patent Title: Pre-Wakeword Speech Processing
Publication date: 2019-05-23
Filing date: 2019-01-24
Inventors: Kurt Wesley Piersol; Gabriel Beddingfield
Assignee: Amazon Technologies Inc.

US20190156818