Voice training data has been ignored when it comes to AI evaluation
Israel Palacio/Unsplash
Artificial intelligence models that generate speech are trained on potentially biased, offensive language, and copyright-infringing datasets, raising concerns about their use.
Generative audio products such as song generators, voice cloning tools, and transcription services are becoming increasingly popular, but while text and image generators have been under intense scrutiny, audio has received less attention. .
To fix this, William Agnew at Carnegie Mellon University in Pennsylvania, and his…
Source: www.newscientist.com
