We’ve trained an AI system that can produce hyper realistic, context-aware audio from text. It is trained on tens of millions of units of data, and is capable of powerful things like copying voices, speaking styles, accents, expressions, volume, pacing and a lot of other parameters by listening to just few seconds of someone’s voice.

We are still exploring its capabilities & limitations with a select group of users to ensure safe consumption of this technology, and are putting systems in place to safeguard against malicious use.