The explosion of neural networks has typically resulted in visual output, with machines creating images based on whatever input the programmers decide, but the Lyrebird program shows the potential that those programs have for audio output as well. Lyrebird is able to create a digital copy of users' voices using just one minute of sample audio, allowing them to then create completely original audio recordings that mimic the user's tone, timbre, and cadence.
To be sure, the ability to copy a user's voice with just a minute of audio is an impressive feat on its own. However, Lyrebird stresses that the program requires a minute of audio at minimum -- longer samples lead to better and more convincing original output.
As proof of concept, Lyrebird has created digital versions of well-known voices like Barack Obama and Donald Trump.