Modulating Custom Voices

360-degree Video
March 10, 2020
A.I. at the Edge
March 10, 2020

Modulating Custom Voices

Generative algorithms are creating synthetic voices that sound just like the original, and those voices can be modulated to the exact pitch and tone desired.

Generative algorithms are creating synthetic voices that sound just like the original, and those voices can be modulated to the exact pitch and tone desired.

Montreal-based A.I. startup Lyrebird built a voice imitation algorithm capable of generating fake speech that sounds indistinguishable from the real thing. It uses a database of voice samples that are either available in public repositories (YouTube, Vimeo, SoundCloud) or samples uploaded by the user. The A.I. learns over time to recognize not only intonation, but also emotional cadences.

This technology can be fun to play with. You can fake a conversation between yourself and your favorite celebrity, provided there are enough publicly-available audio files of that celebrity to build a dataset. Soon, the technology will be able to match and rapidly deploy synthetic voices personalized for each individual consumer.

Y Combinator-backed Voicery creates bespoke voices for brands. If you loved Daria as a kid, you might hear Daria Morgendorffer’s voice (or Jane Lane’s) during a car commercial, while your spouse might instead hear Phil and Lil’s mom from Rugrats.

We should remember that in this era of misinformation, synthetic voices might also be used to trick unwitting consumers.