I made a bunch of these and have some more in the hopper: in these tracks, an amateur computer singer tries earnestly to approximate a real pop star in a casual setting with friends and drinks but does not do a great job.

To make an AI voice model sing karaoke for you:

* Extract a vocal using UVR, or find an a capella, and feed it to an AI tool (I used ACE Studio) that can transcribe the melody into MIDI and write down the lyrics.

* Import the lyrics and extracted melody into another AI tool (or a new ACE Studio project) that sings the MIDI and the lyrics with a generative voice.

* Do a bunch of cleanup on the interpreted melody and the lyrics.

* Build a new backing track from scratch (optional).

* Put the voice (the lime) into the instrumental (the coconut).

* …

* Profit

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit djic.substack.com/subscribe