i'm not sure how that guy did his analysis, but i got a pretty consistent a=0.7 for 1/x^a, using matlab's curve fitter. i graphed the spectral densities using the welch averaging method with 2048 segments, 50% overlap, and hamming windows. i also verified that the curve fitter is accurate by generating a pink noise signal from filtered white noise, and testing the curve fitter on it, and it gets decent results with good r^2 and low rmse, etc whatever, i never took stats.
the point is that the original analysis is probably bullshit. this suisei acapella is not well recorded and has some post processing, and the others he used probably do too. most well recorded vocals will somewhat follow a pink noise curve, and most good masters will approximate a pink noise curve with some bumps, regardless of whether or not they mix with pink noise (which by the way is complete bullshit and a waste of time). if anyone comes to you with some math in music bullshit, its bullshit.
