Thus spake Jim Lesurf:
Snipped
The comparions I'm aware of tended to show that with music and
speech, much higher levels of distortion/nonlinearity were required
to be reliably audible. However for the reasons indicated above, this
is often complicated by the user of speakers that themselves have
levels of nonlinearity well above 0.1 percent.

I can remember recording a test tone onto a cassette tape & being alarmed by
the audible harmonics - couldn't hear them with music. Maybe not the best
example but I suspect that those tones had 1% THD (a bit like valve amps
)