Alexa has been able to speak with celebrity voices for a while in the US and in other countries now offers a choice of male or female voices.
But what if you could get Alexa to speak to you using a friend or relative’s voice? During a presentation on AI at its re:MARS conference, Amazon demonstrated exactly this capability, with head scientist for Alexa AI, Rohit Prasad saying that the “human attributes of empathy and affect are key for building trust… which have become more important in these times of the ongoing pandemic when so many of us have lost loved someone we love”.
In the demo, a young boy asks “Alexa, can grandma finish reading me the Wizard of Oz?” to which, in his grandma’s voice, comes the reply “Ok” and then a short passage from the book.
This is achieved using less than 60 seconds of recorded audio of the person speaking, and is able to turn it into a high-quality AI voice that Alexa can use to say anything, not only reading books.
Prasad said nothing else about the feature, so there’s no word on when (or even if) it will be make available for consumers to use.
Depending upon how you feel about it, hearing the voice of a departed family member could be a way to “make their memories last”, as Prasad put it, acknowledging that it certainly can’t eliminate the pain of that loss.
He didn’t, naturally, mention the downsides to the technology which could be open to abuse. It would be simple, for example, to get Alexa to simulate anyone’s voice, which could be used for fraud or a number of other malicious activities.
Amazon isn’t the only company developing a way to clone people’s voices. Only two days ago, Microsoft published a blog on building AI systems responsibly. It stated that for Azure AI’s Custom Neural Voice, it had “restricted customer access to the service, ensured acceptable use cases were proactively defined and established technical guardrails to help ensure the active participation of the speaker when creating a synthetic voice”.
So, it seems unlikely that Amazon will roll out a voice cloning feature any time soon.