Melanie Martinez is a well-known alt-pop artist, singer, and songwriter. She has over 14 million followers on Instagram, 16 million monthly listeners on Spodify, and dozens of songs. As a big fan of Melanie, I am familiar with her style as I have watched and listened to her content over the years. Recently, I came across an AI cover of Melanie singing Cradles. You can listen to it here.
Again, Melanie did not sing this song, but it sure sounds like she did. The AI system took the original song and made it sound like Melanie was singing. This was likely done by analyzing Melanie's songs, understanding her voice and techniques, and creating a result that matched her artistry as well as that of the song. The AI did a great job with this and produced a very successful result. In fact, it is so realistic that the publisher had to state multiple times that it is not real.
The voice model was created by AIVERSE, which I had not known or heard about about previousely. I did a bit of research on it and learned AIVERSE is a program that builds synthetic datasets to train AI faster.
There are clearly many tools on this platform. Users would start by creating a project and specifying details like requirements for the scene and a call to action. It appears like this AI is used mostly for creating 3D environments. Users receive over 10 different labels including RGB images, 2D and 3D bounding boxes, instance, and classes. See below image that specifies what each of these mean and how they look in a scene.

Overall, this was a very cool program to investigate. It is clearly useful for a wide range of applications as the voice model for the Cradles cover was created using this program. I would like to see this AI become more well-known and am interested to see how it continues to grow.


Comments
Post a Comment