Would you mind elaborating a little or pointing us toward where we can read more about this?

I just (more-or-less blindly, not grokking the actual math behind it) ported the yin algorithm from someone else's implementation this year and tau was defined as a range where tau min is samplerate / freq_max and tau max is samplerate / freq_min where freq_min and freq_max are the bounds of the detection algorithm, at least as I understood it. My port works but I never really understood tau (except as I described) -- if there's a way I can refactor this with a fixed tau that would be very interesting!