A patent application reveals that Amazon is working on a feature that records and identifies the words that a user mentions before they mention the trigger word Alexa.
People who own smart home devices powered by Amazon's virtual assistant have to first trigger the product by uttering the wake word Alexa, which then commands the product to listen to the instruction from the user.
However, in a recent patent application that the online retail giant filed to the United States Patent and Trademark Office, devices powered by Alexa may also begin recording the phrases or words that the person mentions a few seconds before the trigger word "Alexa."
In an example provided by the company in its patent application, users who want to play a song from the Beatles can only trigger their smart home device by uttering an instruction after the word Alexa. Any command before the word will not be recorded and recognized by the personal assistant.
A problem with this approach, Amazon claims in its patent application, is that for some users, this structure may not be natural, and they may prefer providing instructions before the wakeword. To cater to this set of users, the retail giant developed an approach that would record words before mentioning the word Alexa.
The device will pinpoint where the subject of the conversation changes using indicators like lengthy pauses within 10 to 30 seconds before the wakeword Alexa. From this point, the device will send the recording to a server, which will assess if the statement from this marker until the name Alexa is a part of the command from the user.
Amazon claims that this approach should satisfy users who worry about their privacy. The online retail giant recognizes in their patent that recording speech may pose privacy issues not only due to storing conversation data online but also the fact that information will have to travel between the user and the server.
By limiting the speech recorded by the device to 10 to 30 seconds before the word Alexa, Amazon aims to reduce the privacy risk by only focusing on the conversation that could be an actual part of the instruction. However, there is still room for errors, and the few seconds before the direction may contain sensitive information.
This approach may also not be received well by many consumers, especially since a recent survey shows that data security and the possibility of smart speakers and other devices snooping into conversations remain key concerns by customers who use virtual assistants.
Furthermore, recent reports regarding privacy issues have cast doubt on the ability of Amazon to protect user's privacy, and this patent may further exacerbate the concerns of users related to their privacy. For example, a report showed that artificial intelligence researchers could listen to people's conversations stored on the servers of the online retail giant, and if this patent was already in effect, these researchers might have gained access to the few seconds before the user uttered the wakeword.
Moving forward, while this approach has benefits for a specific group of users, and there are attempts to limit possible privacy risks, this patent will surely alert privacy advocates.