My initial thought was that I was going to disagree with this heavily, but it became clear that the specialization was on a different axis than I first expected.

First, let me preface that I think the natural path for almost any data scientist starting their career is to begin as a generalist - create a span as wide as you can until you learn what you care about...

...and then slowly, as you get more experience, try to hone in on explicitly what you want to work on. Here I expected that your "specialization" referred to a certain set of skills, which can sometimes be the case, but instead it seemed more focused on domain (although I think audio is unique on both the domain axis and techniques axis).

While I'm not sure I'd make as clean a statement as "specialize or die", I might say "specializing is the optimal way to move forward quickly, but you limit yourself in domain".