Interesting paper.

I've got a couple of thoughts.

The main one is: What users must BiaB target in order for PG Music to continue to be a viable company?

I suspect that the primary BiaB user is an older user who is interested in reproducing tried-and-true musical styles.

While the price of BiaB is reasonable, it's not exactly inexpensive. A user who wants to create backing tracks will quickly discover the version they want is the one with all the RealTracks. That's likely to be someone with disposable income.

On the other hand, if someone is just looking for something to practice with, a more minimal version of BiaB is an excellent fit.

I'm not well acquainted with contemporary music, but it doesn't seem a niche that BiaB is well-suited to fill. My oldest son is the only one interested in creating his own music, and he shows zero interest in using BiaB. Then again, he's not really into computer-based music creation, preferring to record live instruments.

So I can't really say what sort of market opportunity BiaB might be missing out on by not catering to other groups. But certainly, there's a need to make sure that new users continue to come on board, or they will be faced with an ever-shrinking user base.

There's also a need to maintain the current user base. Certainly, there's a lot of cruft in BiaB. The Mac version seems to be continually behind the PC version. Too much change to the base product - even streamlining existing functions - could lose a portion of the current users, with little gain in new users.

As I see it, AI presents the largest opportunities and threats to BiaB. At the moment, most AI programs seem to be targeted at minimizing the amount of effort by the users, while giving the maximum result. I doubt this target audience would be interested in using BiaB.

On the other hand, these AI programs are capable of creating much more targeted results. For example, they could certainly be trained to output an instrumental track in a particular style that follows a given chord sequence. That's a direct threat to BiaB.

While the current quality of AI tracks is fairly low, I believe this is a function of how the stems are ripped from the source material. But I've heard RVC vocals replace vocals on a Beach Boys track, where the resulting vocals were higher quality than the original recording.

Certainly, AI can be trained to up-sample instruments.

Better funded companies are likely to have access to a massive library of music, and thus capable of creating high-quality custom stems for an end user matching virtually any style in their catalog.

I've suggested before that BiaB could train AI on its own tracks. In theory, this could allow more variety and flexibility to come from existing tracks. In theory, it could also be used as the basis of converting MIDI to audio performance. But while being able to shape the performance at the note level would be useful (for example, to render a specific motif), I'd prefer to maintain a looser level of control. For me, part of the fun of BiaB is treating RealTracks a collaborator. So being able to specify tags at the point of regeneration ("less busy", for example) is something that I'd like.

After all, if I wanted MIDI level control, there are already many excellent VSTi libraries that do exactly that.

But again, being able to supply an excellent product to their base is what BiaB is all about. What's the point of offering rap if very few rappers would use BiaB? That's doubly so when other AI programs will soon be able to offer similar services.

-- David Cuny
My virtual singer development blog

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