Artificial voices

Descript has a neat feature where you can create an artificial voice by reciting a simple script.

You provide a sample of a voice, and Descript uses AI to generate a realistic synthetic version. This allows you to edit audio by simply typing text, and the AI will read it in your voice.

It’s not perfect. Yet. It doesn't always capture every nuance perfectly, and sometimes the generated voice can miss certain inflections or sound a bit robotic.

The audio generated by Google’s new NotebookLM “audio overview” feature--which features two AI generated voices, one male and one female, who explain concepts in a podcast-esque dialog–is, however, astonishingly realistic.

Here’s a five-minute clip of the pair reciting Descript’s voice training statement, and then explaining why the vocal exercise is crucial for mastering natural speech.

It’s scary good. And it raises a lot of questions about what happens when a synthetic voice sounds exactly like a human.

Today is the worst that AI will ever be. We need to be cognizant of the challenges it brings as we embrace the potential that tomorrow’s AI holds.