In the same way that programs like SynthesizerV have replaced concatenative synthesis with AI to generate vocals, BiaB will hopefully offer the option to generate AI tracks trained on RealTracks. Based on what I've seen, this is something that is available with today's technology, not five years out.

AI can be trained using existing tracks on two elements:

  • Learning the note choices an artist makes over chord changes, and
  • Learning how an artist converts notes to audio performances.


Instead of supplying RealTracks as audio played over a given chord progression, RealTracks will become neural networks that can:

  • Convert chord progressions into MIDI notes based on an artist's style, and
  • Convert MIDI notes into audio, based on an artist's performance


This will allow BiaB to do many things that currently can't be done:

  • Training will include performances in all keys. A riff that might have only been available in D would now be available in all keys.
  • AI can interpolate new riffs using notes that didn't exist in the recorded performances.
  • Note choices by one artist can combined with audio performances by another.
  • Users will be able to edit pitch and duration notes directly (like MIDI) and get RealTrack quality audio.


I've pretty much given up on my own vocal synthesis program, because cheap AI vocal synthesis is about to become ubiquitous. It's been a fun journey, but there seems little point in continuing down that path.

I think that BiaB is in an interesting position. They've currently got boatloads of high-quality audio on which to train AI with, assuming the contract allows it. The main problem with AI synthesis is speed - it can take a long time to generate tracks! I suspect that GPUs will soon make that a non-issue for most people.

Plus, AI tracks could be optional. That is, the standard fast method could still be used to generate audio by default. Users could then choose tracks or portions of tracks where they would want to work with using AI.

What current music AI programs don't give people is a level of control, so the appeal is primarily to people who want AI to do all the work. I suspect the audience for that - once the shiny, new aspect has worn out - is smaller than the companies are betting on. Social media will be flooded by sound-alike AI songs, and people will turn to trusted sources to provide curated music.

In contrast, BiaB appeals to people who want that level of control. AI will allow people to have control of elements that had been immutable, unless they used time-stretching and pitch correction software.

-- David Cuny
My virtual singer development blog

Vocal control, you say. Never heard of it. Is that some kind of ProTools thing?