What if your connected speaker was able to speak to you with the voice of anyone, even a deceased person? This is what Amazon wants to offer thanks to its latest advances in artificial intelligence.
Amazon intends to make its connected speakers real members of your household in every way possible. At its Re:Mars conference dedicated to innovations in artificial intelligence and machine learning, the e-commerce giant demonstrated a feature that is both surprising and disturbing: the possibility for speakers connected brand to imitate the voice of anyone, even people who are no longer of this world.
Only one minute of recording required
With less than a minute of audio recording, Amazon’s artificial intelligence would be able to mimic anyone’s tone, intonation and rate of voice. Little information has been given on the operation of this tool, relates TechCrunchbut in a short video shown at the conference, a child could be seen asking his pregnant Alexa if “Can Grandma finish reading The Wizard of Oz?” Immediately afterwards, the speaker changed its voice to adopt a much more human one, close (one imagines) to that of the mentioned grandmother.
According to Rohit Prasad, the chief engineer of the Alexa/Amazon Echo division, the objective of this function would be to “to make memories last […] after so many of us have lost someone”, in reference to the Covid-19 pandemic. Recognizing that “AI can’t eliminate the pain of loss”the official believes that the presence of a familiar voice can help preserve the memory of a deceased person. “We are unquestionably living in the golden age of AI, where our sci-fi dreams are coming true”added Rohit Prasad.
A nest of deepfakes?
Deploying such functionality may raise some ethical issues. Certainly, the feat of synthesizing a human voice with less than a minute of audio is impressive, but in the context of the proliferation of fake news, having an AI capable of mimicking anyone in the middle of the living room could push unscrupulous people to create fake news from scratch, why not accompanied by deepfake videos.
For the moment, the voice replication feature has not been formalized. No date of availability for the general public has been shared by Amazon and we suspect that the tool will still have to prove itself in the company’s labs before arriving on speakers connected to the four corners of the globe.