Amazon is developing a feature that would let Alexa, its in-house voice assistant, mimic any voice after hearing it for less than a minute. The company hasn’t announced a release date yet but did show a video demonstration of Alexa impersonating a child’s grandmother’s voice.
The feature was announced by Rohit Prasad, an Amazon senior VP, at a Wednesday conference held in Las Vegas. According to Prasad, the goal here is to make memories last, adding that many of us lost our loved ones during the pandemic.
Prasad added that Amazon’s aim for Alexa is for the assistant to have “generalisable intelligence”, the ability to adapt to different environments and learn new things with minimal input.
General chatter around AI has been relatively active lately as well, with a Google engineer claiming that the company’s LaMDA AI had gone sentient. Google hit back, denying any such claims and has since suspended the engineer. Prasad made it clear that Alexa isn’t on track to becoming an all-knowing and all-capable AI system, something companies like OpenAI and DeepMind hope to achieve.
However, in addition to the usual chatter about AI, this new feature from Amazon also dives into an otherwise controversial technology. Deepfakes have been both praised and criticised for their benefits and flaws. Scammers have been known to create deepfakes of influential people and use said deepfakes to promote their scams further.
Deepfakes go beyond just scams, though. Microsoft recently restricted which businesses can use its products to mirror voices. This was done to balance the benefit of the company’s voice mimicry tech to help restore impaired people’s speech against the risks of it being used for malicious purposes, including political deepfakes.
In the News: Nothing Phone (1) won’t launch in USA and Canada