Amazon Reveals Alexa Option To Mimic Any Person’s Voice

Amazon has demonstrated its Alexa voice assistant reading a bedtime story in the voice a boy’s deceased grandmother.

The demonstration of the ability to revive a dead loved ones’ voice came on Wednesday at Amazon’s Re:Mars conference in Las Vegas, when Rohit Prasad, senior VP and head scientist for the Alexa AI team, detailed a feature that allows the voice assistant to replicate a specific human voice.

Alexa of course is perhaps the world’s most widely used smart voice assistant, thanks to its tight integration into Amazon’s ubiquitous Echo devices, found in millions of people’s homes.

Generalised intelligence

Rohit Prasad revealed that Alexa receives a billion requests per week, from hundreds of millions of Alexa-enabled devices across 70 countries, speaking 17 languages.

He revealed that more than 30 percent of smart home actions are initiated by Alexa, as it learns routines and habits and predicts them.

Alexa has more than 30 machine learning decision systems embedded within it, Prasad revealed.

But he said customers are demanding more, and this has to be met with “general intelligence”.

This means the AI should generalise its learning, and adapt to the user environment.

This would give Alexa for example the ability learn new concepts with little external input.

Voice mimic

In the demonstration of Wednesday, Prasad revealed that Amazon is developing a system to let Alexa mimic any voice after hearing less than a minute of audio, rather than hours of recording in a studio.

The goal is to “make the memories last” after “so many of us have lost someone we love” during the pandemic, Prasad said.

This could be a real boon to those people dealing with bereavement of loved family members.

Amazon shared its vision for companionship with Alexa at the conference. In a video segment, it portrayed a child who asked, “Alexa, can grandma finish reading me the Wizard of Oz?”

A moment later, Alexa affirmed the command and changed her voice thanks to a personal voice filter.

She spoke soothingly, less robotically, ostensibly sounding like the individual’s grandmother in real life.

Silicon UK readers wishing to see the demonstration hosted by Rohit Prasad in action should skip to the 1:02 hour mark.

Video of Amazon’s Re:Mars conference in Las Vegas

Prasad said “the way we made it happen was by framing the problem as a voice conversation task, and not a speech generation task. We are unquestionably living in the golden era of AI, where our dreams and sci fi are becoming a reality.”

Amazon declined to share when it would roll out such a feature, but the move could potentially run into some opposition, over the risks of potential abuses (including deepfakes) of this touted feature.

Alexa challenges

In December Amazon reacted quickly after its Alexa voice assistant suggested to a child that they do an incredibly dangerous challenge.

Amazon reportedly updated Alexa after the voice assistant “challenged” a 10-year-old girl in the US to touch a coin to the prongs of a half-inserted plug.

The suggestion came after the girl reportedly asked Alexa for a “challenge to do”.

Alexa found the challenge on the Internet and suggested it to the child.

Amazon’s latest moves with Alexa comes after a Google engineer earlier this month was suspended when he made a highly contested claim that a company chat bot had advanced to sentience.