Originally Posted by benwaggoner
By perceptual lossless, I generally mean:
For a home sapian with optimal hearing (no bats!)
Using an optimal auido playback environment (price is no limit)
Using the most challenging real world content (no test signals)
Basically, you're not perceptually lossless if anyone can listen to it and successfully pick out the audio artifacts.
I don't claim to be the optimal homo sapian myself - I'm generally happy with MP3 above 192 KHz. There's plenty at 128 Kbps that makes me want to claw my eardrums out, though.
I'm much more picky on the visual side - I haven't been able to watch VHS for fun since the early 90's. And generally can't even bear to watch TV on hotel TV's due to noise and poor calibration (hence me posting on AVS at this very moment, happily listening to 192 Kbps WMA files :)).
I assume you mean that this perceptual lossless will be with a sample that the subject is intimately familiar with right? Because if not, then noone can know which is the original. In a sense, the perceptual lossless side of things for most homo sapiens doesn't have to be near optimal for either video or audio --- us consumers don't even have the original, so how can we tell whether a compressed presentation is truly perceptual lossless or not?
At best we can say that it is perceptually artifactless, and this is the extent of the consumers' experience unless they happen to have another version that is closer to the original.
My point is that all subjective tests for perceptual lossless as a bar to present audio and video makes sense from economics and efficiency, but from known tools today, it's bunk from presenting the true content itself because the bar itself is movable, and not based on any science, only by subjective tests of ears and eyes based on the compressionists judgment.
Because this bar can move, who's to say that there is any point in changing from 16bit to 24 bit sampling? Who can really tell between 96kHz and 48kHz? For all we know 16bit 48kHz PCM is perceptually lossless from the original 24bit 96kHz. So why bother doing this? Isn't it because the economic and efficiency measure in all of this is suspect and because we recognize that though we crossed the point of diminishing return in accuracy, there is STILL the need to be as close to the original as possible?
As for video, I have even bigger objection to terms like transparency to master or perceptually lossless, not just because of the reasons above, but simply that too many who work in this field and have the golden eyes can take their favorite stress samples and can still pick out the original from any 28Mbps PEAK VC-1 encode of their favorite test samples.
It might be transparent to me, but it is not transparent to them, so unless we agree that the limits in peaks and bandwidths are for economic reasons targetting 90% of the population is enough, then the term is misleading, because the general public now has the perception that the movies presented in the advanced codecs are like PKZIP or RAR when in reality they have fairly large Qp values and are nowhere near lossless.Lossless
is the optimal, economics drives us to strive for bars with qualifiers in front, but since these bars are still subjective, I would say they can't be used because they mislead.
Come to think of it, even the term "artifactless" is false, because while we are all familiar with MPEG2 artifacts (because it is very easy to pick out), AVC and VC-1 hides artifacts very well, and it will take a while for the industry working with the encoding parameters before we can pick them out easily.