I don't think music is 'special' - it's just much more complex than standard test tones. As I said previously, the component sine tones of music are far greater in number, and far more varied in amplitude, and unequally-spaced in frequency. None of the standard test tones are representative of this complexity. The use of simple test tone distortion to judge performance in real-world usage contains the implicit assumption that this distortion has a positive monotonic correlation with the distortion DUTs would produce with much more complex superpositions of many different sine tones. Is there any hard scientific evidence that this will universally hold for all DUTs (or at least just amps)? If there isn't, then the possibility remains this correlation may not be monotonic. And yes, Serge's findings that Df for music (real or simulated) signals does not have a monotonic correlation with Df for sine signals is, while maybe not robust evidence, at least suggestive that the previously mentioned assumption may not be correct. It may turn out none of this applies to amps - I was (perhaps erroneously) inferring that if it's a possibility for DAPs/DACs, it could be a possibility for amps. Nevertheless, as
@waynel pointed out
here, although an ideal amplifier is memoryless, in practice this may not be the case, which could result in their transfer function (and so distortion) being time- and previous state-dependent. In turn maybe these states could be signal-dependent (varying with complexity for example), which could result in an increase in distortion with more complex signals such as music. Who knows. And I think that's the point - how do we know 100% that the mechanisms of nonlinearity we observe with simple test tones are the only ones, and there aren't in fact any others that may manifest themselves only with more complex signals?