Amazon is holding its annual re:Mars conference about machine learning and AI, and has used the opportunity to talk about a confronting piece of technology it's developing for its voice assistant Alexa.

For lack of a better label, it demonstrated what it would be like if Alexa could mimic people's voices so that its speech sounded like your own or a relative's.

The idea was presented by Alexa senior vice-president Rohit Prasad, alongside a short clip of a child being read to by Alexa in the voice of a recently-deceased grandparent.

Depending on your attitude and gut feeling, that might be heartening and sound like a great tool to help ease people through grief, or it could feel creepy and like a step toward a generation that can't distinguish between real life and a voice assistant.

According to Engadget, Amazon says the skill can work to create a voiceprint using as little as one minute of audio of the person in question speaking, so you don't even need a large sample in theory.

It's worth clarifying that Amazon hasn't said this skill will actually make it to users or put any sort of a timeline on a release, but the fact that it's done enough work to show off the tech suggests that it might end up being more than just a demo.