Several have been posted here - sites that host a high quality recording, play it back in trials starting with high distortion, incrementally reducing with each trial. Search and you will find them, and try it yourself if you like. However, in testing trained listeners, audible distortion threshold is generally found to be around 3% in the bass, 1% in the mids and a bit lower in the treble. These distortion thresholds are reflected in the ITU standard for audio equipment used for blind testing (BS.1116, I believe).
The ideal confidence level of an ABX test depends on the goal of the test. An ABX test cannot have both high precision and high recall. High confidence percentiles are high precision but low recall; they reduce false positives but they increase false negatives. Low confidence percentiles (still above 50% of course) are low precision but high recall; they reduce false negatives but they increase false positives. This has been discussed at length on this forum, search and you'll find it.
Do you have a source for studies that have shown trained listeners can detect those levels of distortion? And at what confidence level? I would have thought the levels in the ITU standard for a reference system would be well below audible levels in order to err on the side of caution. You didn't say what your just audible thresholds and confidence level were when you did your ABX tests? As you are claiming to be able to hear distortion differences from positive gain when EQing different headphones (ignoring the fact that negative preamp gain can be applied to partially negate this), the onus is on you to show this audibility with a high confidence-level ABX result that has very low chance of false positives (which would diminish your claim). If you can't do this, then your claim has no scientific merit. And an important distinction - high confidence levels do not 'increase false negatives' (or decrease false positives). It's the probability of these occurring that are increased and decreased respectively. In this case you have made a positive claim of audibility which needs to be backed up by positive evidence with a high confidence level - the fact that the chance of false negatives is higher here is not relevant to your claim. I understand ABX tests are not perfect, but they're the best tests we have at determining actual audible thresholds. If you really want to eliminate false negatives, one potential test would be a null listening test - take your distorted track waveform, minus the original undistorted track's waveform, then ABX this difference track against one of pure silence. This would require no short-term memory, and reduces a complex relative audible difference test to one purely of absolute audible SPL threshold.
EDIT: On second thought, that last idea would probably be too easy of a test to pass, as it doesn't consider the effective perceptual auditory masking of the distortion when the actual distorted track is listened to normally. Best to stick with standard ABX tests, which have proven to be our most reliable tool for investigating psychoacoustics for decades.
Last edited: