Adobe Voco

As the demo showed, the software takes approximately 20 minutes of the desired target's speech and then generated sound-alike voice with even phonemes that were not present in the target example material. Adobe has stated Voco will lower the cost of audio production.[1] With the introduction of Adobe Voco and the similarly capable WaveNet, produced by DeepMind.[2]

Ethical and security concerns have been raised over the ability to alter an audio recording to include words and phrases the original speaker never spoke, and the potential risk to voiceprint biometrics.[1]

There are also concerns that it may be used in conjunction with:

• Human image synthesis, which has reached such levels of likeness since the early 2000s that distinguishing between a human recorded with a camera and a simulation of a human is very difficult.[3]
• Video manipulation of a person's facial expressions in near real-time using an existing 2D RGB video of them.[4]

Adobe's lack of publicized progress has opened opportunities for other companies to build alternative products to VOCO, such as LyreBird.[5]

WaveNet is a similar but open-source research project at London-based artificial intelligence firm DeepMind, developed independently around the same time as Adobe Voco.