AI Daily - Artificial Intelligence News

View Original

AI Can Recreate Singers' Voices To Your Lyrics

OpenAI is a research lab based in San Francisco, their core mission is ensuring that humanity benefits from the development of AI.

Automatic music dates back to more than 50 years ago and OpenAI with a goal to recreate high fidelity music,recently introduced JukeBox a neural net that generates music which involves basic sound reproduction of famous singers in a variety of recreated genres and styles. To train their dataset they scraped and generated a new dataset of 1.2million songs along with the corresponding lyrics and metadata. They then proceeded to train on the 32-bit raw audio and perform data augmentation by randomly downmixing the right and left channels to produce mono audio. This neural net comes well developed, it could track dependencies in repeating melodies.This allows you to re-create the voices of some of your famous singers with your own lyrics.

However there are certain limitations to JukeBox such as there is a distinct gap between generated music and human-created music. The generated songs do provide coherence and can even do amazing solos, we can’t seem to hear large musical structures such as choruses that repeat, and for you international fans, they unfortunately only train on English lyrics on mostly western music but they do hope to include songs from other languages from around the world. For the future their audio team is continuing to work on generating audio samples conditioned on different kinds of priming information. While we are far from listening to AI generated music, OpenAI sure does give musicians a run for their money.