@Guinnless I agree with your point of view, I feel that 24 bit is more important than the sample rate. Given the explanation in the video, it’s pretty obvious that there’s less noise at 24 bit, therefore the audio will be clearer, more precise and airy. So the video simply explains what’s the reason for better audio playback with Hi-Res material, at least from a scientifical point of view. It helps to put things into the right perspective. Other than that we could argue that CD and Hi-Res masters aren’t always the same and it’s not so easy to know which one is better without listening to them.
1 Like
I’ve been reading about 32 bit FP recorders lately and for sure the “almost infinite dynamic range” aspect is the focus. Many don’t even have input level controls for user adjustments. There are good examples of successfully raising the level (in post) of samples taken at very low gain, but still so far above the noise floor that they are unaffected. Similarly, examples of lowering the level of samples that look (on a 16 or 24 bit view) to be way over saturated and clipped, but remain well within the digital dynamic range.
I’ve not been able to reliably identify hi res vs redbook using the Dr Aix blind test files of identically mastered source material… well recorded 16/44 is good stuff. But it definitely seems easier to make well recorded 24/x or especially, now, 32FP/x recordings and work from there.
In fairness, though, a slight weakness of this test video is that it seems to present analysis of a 1kHz sine wave (I haven’t watched)… and that is such a simple case that almost any sampling rate - and certainly 44, 48, 96 or 192 kHz will not be a limiting factor (there being, indeed, no stair step). Indeed, I’d say that it is constructed specifically as a test of dynamic range or bit depth, hence noise floor.
Cool beans.
In the video the test is conducted changing the Hz value from 1 Hz to 20.000 Hz.
Thanks, I will take a look for sure… but the (obscure!) point I was making is that there is zero doubt about reconstructing a single frequency sine wave (a la tone generators) in almost any DAC which obeys Nyquist (2x sampling rate vs highest frequency of interest). Noise floor (from bit depth) is the main thing under test for this kind of example I think.
The “tricky” part - which may lead to hearing more airiness, details of timbre, and maybe even spatial location - is almost certainly related to phase (stereo) and/or waveform shape (mono frequencies, but high harmonics like square waves) differences that may be related to much higher frequencies of interest and therefore sampling rates.
Does that make sense? Thanks for the follow up.
A good, clearly-explained video. I like the comparison with cassette tape noise as being the equivalent of recording at just 6-bit depth.
16/44.1 is great. It was great in the 80s (when it was done well) and is great today (ditto).
Mark
1 Like
Found it, enjoyed it… and I recall seeing the original source video (with the bearded guy and the spectrum analyzer) that he used from his classroom training… I’m old I guess!
Really nice summary and a great reminder that the “lollipop” graph of samples and the “zero hold” continuous representation are not the same thing, and that either one is only half way back to the reproduction of an analog signal. The Nyquist Theorem isn’t just a “theory”, it’s a fact - and (for me the quote of the day) the stair steps were never there.
Thanks for this, great intro to a topic close to all of our hearts.
1 Like
@alan33 yes it makes sense, even though I’m not aware of all the technical aspects of the topic. However in that video they do compare different sampling rates if I’m not mistaken and the output sound wave appears to be identical. Before watching it, I thought higher sampling rates were the reason for airiness and detail, however it seems that bit depth is the real reason for that. I’m no expert, but it seems convincing to me.
I’m glad you @Ebor @alan33 enjoyed watching it, I think it’s an interesting topic of discussion given that we all strive for better audio playback in the end and we need to choose which format to use. So far CDs have given me an excellent experience, while Hi-Res hasn’t often lived up to its standards.
1 Like
The catch here is that if there is no frequency content in the input signal above 1/2 the sampling frequency, they all will be identical. That’s what the Theorem is saying.
If you only wanted to preserve 1 kHz signals, you could use a 2 kHz sampling rate… but that isn’t what we want for audio. They didn’t show what “goes wrong” when the input signal has meaningful content above the sampling frequency cutoff.
If the input frequencies that lead to more airiness or detail are higher than 22.05 kHz (half of Redbook’s 44.1) - hence ultrasonic for our ears - then they will only be preserved if the digitization process uses higher frequencies.
That is the basic assumption being challenged: Redbook says nothing above 22.05 matters; hearing better (or even different) sound when using 96 or 192 says otherwise.
The only part of the video that (almost) addresses this is where the bearded guy is explaining that there is only one unique way to draw a smooth line through the lollipops, and then he shows a slightly distorted different curve and explains that this one would have had content above the Nyquist cutoff frequency… it’s a complicated point to unpack.
As I say, I got close to 50% when I did the Dr Aix test on a couple of dozen Redbook vs 24/96 files, so the ultrasonic content doesn’t make a difference to me: identical masters sound the same to me once you get to Redbook. But for others, maybe it does.
But the advantages in the improved recording chain to make a better mix down, even to “only” Redbook for guys like me, certainly are real and discernible (as the Audio University guy mentions at the end).
I hope this long tedious post is okay: I didn’t want you to take away a wrong interpretation about the S/N advantages being the only possible advantage. There’s lots to consider in the frequency domain (sampling rate), not just in the amplitude (bit depth) domain. A classic example being the type of ‘anti aliasing’ filter used to eliminate the (mathematically repeated) content above the sampling frequency: Naim spend a great deal of time optimizing their DSP filtering code, and don’t bother using the built-in filtering of the Burr-Brown / TI DACs that they use.
At this level, everything matters, but it isn’t always important because of the “intuitively obvious reasons”… and that’s a key message from the Audio University presenter!
1 Like
I’m obviously going down a rabbit hole on this excellent video, but that 6-bit claim for audio compact cassette also struck me as weird! I mean, cassettes sounded way better than early computer 8-bit audio beeps and chirps!
A quick refresher on calculating dB from bit depth (thanks Wikipedia) proves this out…
Tapes had SNR of about 60-65 dB, which is about 10 or 11 bits equivalent. OTOH, 6 bits is only about 37 dB - way below what we would need to think of something as good quality audio.
I’m not sure how the bearded guy did his calculation, but 6 bits / 37 dB is almost certainly pessimistic (for dramatic purposes) when comparing against Redbook at 16 bits / 98 dB. Maybe he was using a cheap 1960’s portable recorder, crappy tape, and the crummy built in mic that I first used to compile mixed tapes as a boy?!?
I can only agree about these points. And yes, the topic can be both fascinating and complicated for sure!
Is Dr Aix test still available? I looked for it but couldn’t find it. I would be curious to see my results.
Nothing against Hi-Res audio, I do welcome it indeed. However it becomes meaningful when you can spot a performance difference, otherwise it’s just a technical exercise.
Don’t worry, it’s not a tedious reply in my opinion, it’s just more technical than the average post and it’s nice to have a contructive discussion. As you rightly said, there’s a lot to consider and everything matters in the end. The fact is that there are so many things involved in the end result, that one might get confused when looking for a simple answer. I wasn’t aware of Naim DSP optimization, surely that’s another important key factor. Also one should always consider how the room, the Hi-Fi system and all its components could affect the audio perception. For example some speakers tend to enhance high frequencies, others might not.
In the end I try to understand what’s going on in the audio chain, but I try not to get crazy about it, the important thing is to enjoy music.
True, but firstly, that’s 8-bit sound generation (synthesis) you’re referring to, not sound recording. Secondly (and this is without rewatching it to check), I recall he was just referring to the noise level on cassette tape as being the equivalent of that bit depth, having just pointed out that increasing bit depth serves mainly to shove the noise floor down.
16/44.1 rules. I may have to get a T-shirt made to show how edgy and cool I am when it comes to digital audio.
Mark
1 Like
Thanks! I wondered how far off the deep end I was going for a friendly chat!!
I think it was a limited duration public test as part of his preparation for last year’s audio society conference. He seems to have been offline for a while but I had a bulk email a week or two ago saying that he’s back. Perhaps he still provides the files and compiles score if you email him? I really don’t know. It was very interesting indeed, and I was happy to get to try it!
I raised it almost tangentially, since it has come up a few times in the New Classic thread, where Neil has carefully explained how the (cheaper) TI chip chosen is used in a very limited case fashion, with all of the digital processing happening off-board on the SHARC processor - and in this mode of operation they found the cheaper chip to sound better that the higher cost, better native-spec chip. In part, the better published TI noise spec is related to differences in the way the two chips internally handle digital filtering. The published noise floor differences may also relate to voltage filtering improvements internal to the more expensive chip… but again, Naim manage that themselves via extensive off-chip regulation. These really are system wide optimizations. Once again the key message is that there is more to this than meets the eye, and the “intuitively obvious” explanation may not be anywhere near relevant!
Thanks again for posting this thread and the video. Very fun.
I guess it always depends on the people you deal with. Some might like it, some not.
That’s a good idea, maybe I can get in touch with him. The test is surely very interesting!
I’ve been reading the New Classic thread from time to time, maybe I missed the point you are talking about. I usually don’t care too much about the chip used in the DAC because it’s just half of the story in the audio chain (also some cheap streamers have sometimes better specs than some expensive ones. But it’s just specs on paper of course). What you are saying just proves it. Cheaper doesn’t meant it’s going to be worse. It always depends on the circumstances. I guess you learn all these little things over time and with experience. Our own personal experiences define our ideas and credo indeed.
I’m very glad you enjoyed the video and thread!
1 Like
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.