Now AI can sing and Write songs! - Jukebox from OpenAi made that possible [You can listen ai generated songs!]

The revolution that Artificial Intelligence making in this world is phenomenal!. If you take the number of things that we use AI currently for, i would say its pretty much huge to count. Infact, you might need an another artificial intelligence based program to count that too... The list grows day by day, be it a writing skill, coding skill, security skill or any other skill AI is cementing its place on that, and it's evident because of these new discoveries.

And now an AI singing and writing songs joins that list as well.  Researchers from OpenAi in collaboration with various others has developed an artificial intelligence named Jukebox that can sing your favorite song like the person who had actually sung, when you just entered lyrics! that's it? no, it can even generate own songs.

There exists multiple technologies like ai reading out the text we entered(TTS), generating text based on audio file etc, but this technology is a little bit different and can be used in future to recreate the song you like with just lyrics using your favorite singer's voice without asking your favorite singer who might be alive or dead at that time, No offense! everyone needs to die one day. 


We have introduced Jukebox, a model that generates raw audio music imitating many different styles and artists. We can condition this music on specific artists and genres, and can optionally specify the lyrics for the sample. We laid out the details necessary to train a Hierarchical VQ-VAE to compress the music effectively into tokens. While previous work has generated raw audio music in the 20–30 second range, our model is capable of generating pieces that are multiple minutes long, and with recognizable singing in natural-sounding voices.

It wasn't an easy task particularly the Lyrics to Singing task(LTS) says the developers because of alignment of lyrics , the artists voice and also the diversity of ways one can sing a phrase depending on the pitch, melody, rhythm and even genre of the song.. But finally made this successful model that can do the task specified, after trying and conditioning different set of models. 

Kudos to the developers!

So, i guess this is the time to get ready for the future!

You can read the complete paper here.

You can hear the ai generated songs, in the official page here which has 7100+ songs as of writing this post. 

If you are interested to further develop this or want code of this, then you can get it here

Related Posts