Voice-Swap, an ethically-trained AI voice company, and BMAT Music Innovators, a company that indexes music usage and ownership data using machine learning, have partnered to launch a new technical certification for AI voice and music models. It is designed to verify that the audio content used to train voice models does not infringe on any copyright or publicity rights.
While ex-Stability AI audio executive, Ed Newton-Rex, also recently launched a certification for AI models using licensed datasets through his non-profit Fairly Trained earlier this year, the new certification from BMAT and Voice-Swap has a distinct approach. Fairly Trained is based on voluntary disclosures about what materials are used in AI training using a written form, whereas BMAT and Voice-Swap’s certification is more technologically-based certification.
Companies that wish to earn the Voice-Swap and BMAT certification must allow the two to analyze a given model’s training data and compare it against the 180 million copyrights BMAT has logged in its music index.
BMAT’s CRO, Alex Loscos, says of the partnership: “At BMAT, we’re driven to be pioneers at the intersection of music and technology. Our certification of training datasets is a major step forward –the first of many groundbreaking initiatives we have planned. We’re excited to have found such a like-minded partner in Voice-Swap, who’s been instrumental in helping us bring this vision to life and who share our passion for transparency in music.”
“We believe that AI should be beneficial to artists,” says Voice-Swap CEO Dan “DJ Fresh” Stein. “This next-generation audio-screening certification could be the first step to solving the music industry’s AI problems.”
