Test: 24 bit vs final CD product

[bob katz]

Originally Posted by Key

See I am throwing every test I can at this.

SNIP

But like I said the added noise in dithering to anything higher than that seems to be higher than the distortions incurred.

It's good to see you investigating! Maybe some mysteries are better left unsolved :-).

But seriously, the key word which is missing is the word "audibly". In other words,

"But like I said, is the added noise in dithering to anything higher than that AUDIBLY higher than the distortions incurred?"

And the answer to that question is psychoacoustic, not viewable on any FFT! Many of the sorts of distortions that come from truncating complex calculations result in high frequency distortions that are quite audible even BELOW the noise. I really don't think you should be fooling with the psychoacoustics of dither when it's clear that DITHER HELPS AND IT DOESN'T HURT. Especially at the 24 bit level. At the 16 bit level it can be argued that dither masks some low level information, but the ugliness of the distortion that's revealed when truncating is always worse-sounding than any information which may be masked. At the 8 bit level, many folks feel that no dither or partial dither is preferable to the dither noise, but who makes 8-bit files anymore? Or cares about their quality?

[John Moran]

Originally Posted by PatrikT

Try tests like these with spinning turntables or a rotating tape machines to realize something.

what sort of dither should i use for the turntable? should it be applied directly to the disc or on an insert after the cartridge but before the riaa curve? should different dither be used for the tape machine? 30 ips is bigger than 24 bit so 30 ips should sound better, but 15 ips is smaller so that won't sound as good as 24 bit, right?

[Key]

Originally Posted by bob katz

It's good to see you investigating! Maybe some mysteries are better left unsolved :-)......

......Many of the sorts of distortions that come from truncating complex calculations result in high frequency distortions that are quite audible even BELOW the noise. I really don't think you should be fooling with the psychoacoustics of dither when it's clear that DITHER HELPS AND IT DOESN'T HURT. Especially at the 24 bit level.

Well I am open to the idea that there are just some things we can't "know" or prove. But we can see where the evidence points and make a decent conclusion.

The idea of hearing distortions past the noise floor is where I will differ in opinion with you. I think there has to be a point where distortions are negligible. At least provided that the distortions are not some how congesting the transducers in the signal chain or intermittent. Provided they are constant they are most likely just going to be swamped out by much larger problems in the signal chain - low level distortions from the speakers.

[bob katz]

Originally Posted by Key

Well I am open to the idea that there are just some things we can't "know" or prove. But we can see where the evidence points and make a decent conclusion.

The idea of hearing distortions past the noise floor is where I will differ in opinion with you. I think there has to be a point where distortions are negligible. At least provided that the distortions are not some how congesting the transducers in the signal chain or intermittent. Provided they are constant they are most likely just going to be swamped out by much larger problems in the signal chain - low level distortions from the speakers.

You may be right about that; I'm not going to argue over thresholds as I don't know any better than anyone else when it ceases to be audible; but just be careful about leaping to conclusions about the audibility of low level distortions. First of all, distortion accumulates, and they tend to rear their ugly head (perceptibly) three generations down the road. Secondly, it's better to be conservative than to take this kind of risk, since the audible "cost" of dithering (particularly at 24 bits) is so low that not dithering is just not a chance you should take.

As for being swamped by loudspeaker (or rather amplifier) distortion, I'd like to point out that the two distortion mechanisms are very different. Analog amplifier and loudspeaker distortion tends to be more harmonically-related while quantization distortion is inharmonic as its origins are beating against the sample rate. So one mechanism does not tend to mask the other. In other words, you can detect another type of distortion even through another distorted system.

[Key]

All great points. But I am still unclear on the truncation vs dither at 24-bits.

The thing is in my case 1. I can't hear a difference (yet) 2. The benchmarks and measurements I am taking are actually pointing in the direction of truncation in order to side on caution. But I may be just interpreting the data wrong or doing the wrong type of comparisons. And of course this is only true for 24-bit or higher.