turian
commented
3 years ago "It's not clear to me that a number of transforms, e.g. "bandpass filter," represent a monotonic dimension of variation, taking it quite a distance from the JND paradigm." @maxsolomonhenry Why? If you take one parameter and monotonically increase it, why not? i.e. as width_q and midi increase it should be monotonic?
Also, monotonicity is not required for our experiments.
If pure phase shift is imperceptible, we'll learn that pretty fast from our experiments :)
Also doesn't really matter if chorus and flanger are deterministic or not, given our experimental setup
I'd double check to see that chorus and flanger always start at the same modulation phase. It might be that the same file processed with the same settings will sound different, opportunistically, depending on how SoX manages the modulation cycles.
Re: Allpass, I don't believe a pure phase shift is perceivable, but I'd happily be proven wrong. Maybe it'll monkey with the transients.
It's not clear to me that a number of transforms, e.g. "bandpass filter," represent a monotonic dimension of variation, taking it quite a distance from the JND paradigm. We might reframe this experiment instead as "detection of common audio effects." Which is still likely useful.
https://github.com/turian/audiojnd/blob/76c6b09e72a56184cef28904cfe7e77d086250b7/transforms.py#L28