Quote:
Originally Posted by Roger Dressler /t/1425980/opinions-on-apple-tv-for-streaming-audio#post_22340383
Yes, agreed. But the original statement about upsampling never offering any value was not confined to 48 kHz.
There is objective data, but that does not translate to subjective quality.
Ok, so it was my sweeping generality that you differ with. Fine, I can accept that.
Quote:
Originally Posted by Roger Dressler /t/1425980/opinions-on-apple-tv-for-streaming-audio#post_22340383
And again, I was not implying that the Apple upsampling was to be seen as either an improvement or a degradation, but simply illustrating that we cannot assume it has zero audible side effect simply because most people don't hear it.
But if most people don't hear it, what are we going for here? I've listened to upsampling, I have to assume done right, and remain just a bit unconvinced that it does anything except take up more space in storage. I would not say the same of high rate originals, though. But somebody needs to explain why upsampling, even radically, would improve a 44.1KHz original. The filter argument only goes so far, especially if the original was done in the first decade of digital with analog filters. How can upsampling remove their signature, audible or not? Yes, it changes the final reconstruction filter requirements, but that's comparatively trivial in relation to what's already been recorded into the master.
I'm probably making an error in this comparison, but if you look at upsampling a digital image, you can add pixels that are interpolated between the originals, but you never add image detail. You effectively smooth the transitions between original pixels. However, if a visual reconstruction filter is applied (presenting the image to a viewer at a visual size where the pixels are not visible in the original) the up-sampled version at the same size looks identical to the original. It's only when we magnify the image to the point where original pixels become visible that the upsampled version may look smoother. No more detail, just less jagged edges and transitions. However, the problem with the picture analogy is two fold. First, scaling an image takes into account adjacent pixels in all directions, and in interpolating samples in sampled audio you only have two adjacents to work with (there is also a "sharpening" function that improves edge transition rates, but doesn't add missing detail). Second, we can magnify images, we can't do the same with audio. Turning it up louder isn't the same thing at all.
So in audio, we're left with an original where samples have been smoothed to inaudibility, and an upsampled version where the samples have been smoothed to inaudibility. In neither case do samples themselves ever come out of the reconstruction filter.
Then there's that darn AES paper: "Audibility of a CD-Standard A/D/A Loop Inserted Into High-Resolution Audio Playback" from 2008, which while not directly analyzing upsampling, does take some shots at the whole high-rate concept.
Please tell me what I'm missing, and where the significant improvement could possibly come from upsampling audio. I'd love upsample to make an improvement, I just can't see it happening. Convince me otherwise, please.