[bigshot]

Maybe I'm misunderstanding, math isn't my best subject, but I'm not talking about frequency, I'm talking about how long the attack and decay of a fast snare drum hit would be in music- real world transients, not theoretical ones. I've looked at waveforms of drums and none of the hits were anywhere near the microsecond range.On the fastest hits, it usually took 2 or 3 milliseconds for the attack to register, then a decay that lasted at least ten or twenty times that, often much longer, Go look at the waveform of a drum hit. It isn't like a brick wall hitting all at once. It's a mountain that rises to a peak and then trails off. It seems to me that worrying about anti-aliasing affecting real world musical transients would be like worrying about a rock or two being out of place on Mount Everest.

 

[pinnahertz]

Maybe I'm misunderstanding, math isn't my best subject, but I'm not talking about frequency, I'm talking about how long the attack and decay of a fast snare drum hit would be in music- real world transients, not theoretical ones.

I'm talking real-world music also. Time and Frequency are just a different window on the same information, different "domains" if you like. The two are inseparable, as they are the same data.

I've looked at waveforms of drums and none of the hits were anywhere near the microsecond range.On the fastest hits, it usually took 2 or 3 milliseconds for the attack to register, then a decay that lasted at least ten or twenty times that, often much longer, Go look at the waveform of a drum hit. It isn't like a brick wall hitting all at once. It's a mountain that rises to a peak and then trails off.

What you refer to here is actually the "envelope" of the signal, and your numbers are fine for that. The envelope is the overall amplitude vs time graph of the hit. If you zoom in and look at that waveform a bit more carefully and note the specific rise-times involved at various points, you'll see there are components of that sound that rise much faster than 2-3ms, fast enough to produce frequency components all the way up to 20kHz and beyond. I did an FFT of a single snare hit sample here and saw dense spectrum up to Nyquist. You can't get that if your fastest rise time is 3ms (that inverts to 333Hz). The math is simple, reciprocal of time is frequency, and vise-versa.

It seems to me that worrying about anti-aliasing affecting real world musical transients would be like worrying about a rock or two being out of place on Mount Everest.

Well, perhaps on some level, but aliasing is nasty, folding products way down into the audio spectrum. You'd hate it if you heard it. Less aliasing is better, aliasing can be audible and not harmonically related, whereas phase distortion around filter cutoff is not as easily audible.

 

[pinnahertz]

I understand how this applies to continuous tones, but not how it applies to transients.

Why would it be different?

That "uncorrelated noise" is the error being spread (and resolved) over time - is it not?

No, that's not quite right. One way to look at it is changing a hard quantization threshold to a smooth transition by virtue of uncorrelated noise.

If so, don't we need the error fixing at the time of the transient, rather then over the next few hundred samples?

No, a transient is typically a higher level signal, and wouldn't be affected (or need "error fixing") until it gets way down into the noise floor.

 

[castleofargh]

Jiggling a signal that’s not causing a bit transition pushes it over the transition point for an amount statistically proportional to its actual amplitude level.

My argument is that for short transient sounds in music you haven't got enough samples for those statistics to work.
The correct fix for quantisation error in 16bit audio is therefore to switch to 24bit audio, which today is simple and almost without cost.



Classical music has a wide dynamic range with quiet stuff being played at reasonable volume. With a decent amp and speakers: or decent headphones, this is not an unreasonable scenario in the HiFi world.
You appear to be saying 24 bits is more accurate than we need...
...What's the drawback of using a format that's better than we need?

I do think it is unreasonable in the hifi world. maybe you should get some measurement gears and test what your system, and listening environment do to the music in real life. if you think that dither or any form of added noise is damaging the music, then you need to keep that view and agree that any other and louder noises or distortions are doing more damage and most likely making the dither noise irrelevant because of the difference in magnitude.
but I don't even see audio that way. all I see is one sound + one sound + one sound + some noises, all in one wave per channel. all the sounds are entirely contained(but band limited)in the file. some extra noise isn't changing the instrument, some extra noise is only making extra noise. unless you have so few bits that you start cutting the amplitude of the signal itself, which in this case would require a music piece of more than 90dB of dynamic, quantization noise is only noise. extra noise down at around -90dB. do you hear such noise in normal listening conditions? no. and if you did somehow in a strange and unlikely situation, after dither it would be even less noticeable.

anyway, there is nothing wrong with using 24bit aside from making the file bigger and costing more. if you're fine with that, enjoy. I'm arguing the motivation for using 24bit in a playback system, not the use of 24bit.

 

[gregorio]

[1] My argument is that for short transient sounds in music you haven't got enough samples for those statistics to work....
[1a] That "uncorrelated noise" is the error being spread (and resolved) over time - is it not?
[1b] The correct fix for quantisation error in 16bit audio is therefore to switch to 24bit audio ....
[2] Classical music has a wide dynamic range with quiet stuff being played at reasonable volume. With a decent amp and speakers: or decent headphones, this is not an unreasonable scenario in the HiFi world.
[2a] ...What's the drawback of using a format that's better than we need?

1. Two points: Firstly, as long as the transient contains frequencies with a waveform duration of at least 2 samples then we can encode and dither it perfectly. With CD (16/44.1) that means transient peaks with freq content no higher than 22kHz. Secondly, it doesn't matter what amplitude values we have or how quickly the amplitude changes, providing of course we do not exceed the clipping point (0dBFS).
1a. No, the error is not spread over time. I think you're getting confused by the fact that the quantisation error of each sample is rounded but as each sample is rounded up or down randomly with dither, you obviously need several samples for it to be apparent that the rounding is random (and therefore statistically perfect). Dither operates as well (effectively perfectly) on transient peaks as it does on any other variation of amplitude (within the Nyquist limit) and the dither not appearing random could only potentially be a problem with commercial recordings which only last a few dozen micro-secs and of course, there aren't many of those! BTW, I say "potentially" because in practice you couldn't identify any sort of quantisation error in a recording of such short duration.
1b. No, 24bit does not fix quantization error, it just reduces it. Dither on the other hand does fix quantization error, after dither there is no quantization error at all and therefore, dithered 16bit is more accurate than un-dithered 24bit.
2. Yes, classical music has a very wide dynamic range. Symphony orchestra recordings typically have the widest dynamic range of any music genre, up to a maximum of about 60dB, which is the equivalent of about 10bits ...
2a. Absolutely true and that's what we already have with 16bit! A typical commercial music recording has a dynamic range equivalent to about 8 or fewer bits but it can go as high as about 10bits, as mentioned in #2. 16bits therefore provides over 100 times more dynamic range than is typically used and about 40 times more than is required even for the more extreme symphony orchestra recordings. How much more "better" do you want? Apart from another 8bits of inaudible noise and audio files which are a third larger, what do you think you actually gain from 24bit?

G

 

[Cutestudio]

What kind of transient are you going to find in music that doesn't cover hundreds of samples?

It's a good question, here's a clip that has 93 samples in:

Some of those shapes in the selected area are formed by a very small number of samples.

When I look at that waveform I find that each vertical point is quantized to a particular level, so for instance a distant (quiet) 'click' sound at an average level of -60dB will have a quantisation error of around 1/255 of fsd. This error will be there regardless of dither: imagine you can print 50 samples out onto clear sheet of an 8 bit non dithered, 8 bit dithered, and the analog, and you placed them all on a lightbox to they overlapped perfectly.

Then as a test you could print out the analog waveform with some pure analog white noise of equal amplitude to the dither signal, and lay that on top for a 4th version of that waveform.

My objection to settling for the 'good enough' 16bit over the easily accessible 'better than we need' 24bit is therefore demonstrated by the errors in these waveform shapes.

I am aware that the maths in the continuous case says that it's all just a type of noise, but isn't any distortion a type of noise?
The fact remains that if the aim is to preserve the waveform shape (surely the very definition of 'High Fidelity'?) why are we arguing that 16bit is an acceptable solution when the waveform snapshots clearly show errors for transients in quiet passages?

I'd understand the defence of 16bit more if we'd hit a real issue with purity of iron cores, resistance of silver wire etc but moving from 16 to 24bits? In todays age of 1920x1080 Netflix streaming, huge storage,quad core 1.4GHz phones and smart fridges: 24bit is trivial. In fact we did it two decades ago back when everyone thought Windows 98 was good, with SACD and DVD-A.
https://en.wikipedia.org/wiki/DVD-Audio

 

[gregorio]

[1] When I look at that waveform
[2] so for instance a distant (quiet) 'click' sound at an average level of -60dB will have a quantisation error of around 1/255 of fsd. This error will be there regardless of dither ...
[3] My objection to settling for the 'good enough' 16bit over the easily accessible 'better than we need' 24bit is therefore demonstrated by the errors in these waveform shapes.
[4] I am aware that the maths in the continuous case says that it's all just a type of noise ...
[5] The fact remains that if the aim is to preserve the waveform shape (surely the very definition of 'High Fidelity'?) why are we arguing that 16bit is an acceptable solution when the waveform snapshots clearly show errors for transients in quiet passages?

1. You are not looking at a waveform!!! You are looking at a graphical representation of the data point on your disk, NOT the waveform once it has been converted (via a dithering quantiser in the DAC)!
2. Again, NO! It will have zero error after dither.
3. No one is arguing for "good enough" (10bit audio), we are arguing for at least 40 times "better than we need" which is what we get with 16bit. What do you not understand?
4. Yes, white noise that's 40 times lower than the noise floor of any commercial recording! What is there here which you do not understand?
5. The only fact which remains is that you're refusing to understand how digital audio works and are using any audiophile myth you can find to prove a point about 24bit which is false! The fact is, that with dither the waveform shape is perfect, down to about -92dBFS with dither and then it's still perfect but covered by white noise.

Answer the question please, how much more than 40 to 100+ times more dynamic range than is ever used do you want, and why?

G

 

[bigshot]

What you refer to here is actually the "envelope" of the signal, and your numbers are fine for that. The envelope is the overall amplitude vs time graph of the hit. If you zoom in and look at that waveform a bit more carefully and note the specific rise-times involved at various points, you'll see there are components of that sound that rise much faster than 2-3ms, fast enough to produce frequency components all the way up to 20kHz and beyond. Less aliasing is better, aliasing can be audible and not harmonically related, whereas phase distortion around filter cutoff is not as easily audible.

The envelope of the sound is the sound that we recognize as a drum hit when we hear it. Individual frequency waveforms within that are part of it, but on a very small scale. If a recording medium is able to capture a frequency up to 20kHz even with anti-aliasing, I can't for the life of me see how anti-aliasing could affect transient time in a drum hit to the degree it isn't audibly the same drum hit any more. That's like saying one pebble out of place on Mount Everest changes what Mount Everest is. I think we're saying the same thing, but I'm standing back a bit and looking at the overall sound of a musical transient and you're pulling out a microscope and looking at one sliver of it. I always find that it's easier for me to keep a realistic perspective if I keep focused on the overall stuff that really matters.

My general point here is that the way modern DACs handle the top end, whether it uses a brick wall above 20kHz or a roll off above 20kHz, it doesn't matter because the difference between those two isn't audible. I see you saying basically the same thing. You're just assuming that because there is some theoretical difference in phase distortion, you add that qualifier "not as easily audible". I don't need to do that because the difference is so small it's clearly below the threshold of perception. Our ability to hear phase error is in the millisecond range. The phase error of cutoff filters is well below the threshold of perception. The transients of drum hits in recorded music exist on an even larger scale than that. Therefore, worrying that transients in music might be affected by phase distortion from cutoff filters is essentially running down a mental rabbit hole. Audiophiles of all kinds tend to do that too often because they can't see the forest for the trees.

 

[bigshot]

I just realized I conflated this discussion with the one about slow and fast rolloff. Sorry about that!

It's OK though, because I had the audacity to ask a fella over there if what he was talking about was audible to human ears. He said that question was "derailing the thread". There ya go!

 

[Cutestudio]

1. You are not looking at a waveform!!! You are looking at a graphical representation of the data point on your disk, NOT the waveform once it has been converted (via a dithering quantiser in the DAC)!
2. Again, NO! It will have zero error after dither.
3. No one is arguing for "good enough" (10bit audio), we are arguing for at least 40 times "better than we need" which is what we get with 16bit. What do you not understand?
4. Yes, white noise that's 40 times lower than the noise floor of any commercial recording! What is there here which you do not understand?
5. The only fact which remains is that you're refusing to understand how digital audio works and are using any audiophile myth you can find to prove a point about 24bit which is false! The fact is, that with dither the waveform shape is perfect, down to about -92dBFS with dither and then it's still perfect but covered by white noise.

Answer the question please, how much more than 40 to 100+ times more dynamic range than is ever used do you want, and why?

G

I'm not sure why you are saying I'm not looking at the waveform, do you suggest I get inside the wires and watch the electrons drifting back and forth?

I don't think you realise what dither actually is, it's just noise added before quantisation to even out the quantisation errors, it doesn't actually make each little wavelet have the right shape, each point will still be quantized to the wrong value.

There's a Wikipedia article that's fairly good at explaining:
https://en.wikipedia.org/wiki/Dither

If a series of random numbers between 0.0 and 0.9 (ex: 0.6, 0.1, 0.3, 0.6, 0.9, etc.) are calculated and added to the results of the equation, two times out of ten the result will truncate back to 4 (if 0.0 or 0.1 are added to 4.8) and the rest of the times it will truncate to 5, but each given situation has a random 20% chance of rounding to 4 or 80% chance of rounding to 5. Over the long haul this will result in results that average to 4.8 and a quantization error that is random — or noise.

Note the phrase 'Over the long haul' as a clue to the temporal difficulties: the quantisation is still there, dither is not magic, it just trades the repeating error into noise over time.
This noise is also itself quantised of course and there are various home brew dithers that sound better than others, but it's not analogue noise, it's still stuck to a number of levels like a randomised PWM.
For instance a brief level of 0.5 bit is transformed from a truncated 0 with dither to a random 50/50 split between 0 and 1 which one hears as noise. Not a nice noise as in analog, but a rather coarse noise: try it on an 8 bit waveform and listen.

Why exactly are you against 24bit, do you need a bigger disk or is it metering charges on the internet?

Here's a study that found 24bit was more perfect than 16bit BTW:
http://www.aes.org/tmpFiles/elib/20170620/18296.pdf

But even disregarding that, I'm puzzled by the fight for 'good enough' when clearly 24/96 is not only achievable with ease but it seems used by everyone outside of audio: why have the HiFi crowd got the worst format? Is the obsolete CD format really that important anymore? I can't recall the last time I listened to a CD on a CD player, they arrive in the post and get ripped that day.
Perhaps the lack of easily available iTunes downloads at 96/24 isn't the record companies fault after all, but out fault for fighting and insisting we get sold an inferior format?
Why are we demanding mediocrity in audio? Do we turn down 100W amps because we may only want 2W to use? Is that speaker too good for us so we get a lower grade one? Is that steel beam holding up the house too good for the job and we demand a smaller one? Format wars appear to be the only branch of HiFi where people demand less, rather than more.

 

[pinnahertz]

But even disregarding that, I'm puzzled by the fight for 'good enough' when clearly 24/96 is not only achievable with ease but it seems used by everyone outside of audio: why have the HiFi crowd got the worst format?

Where is 24/96 used outside of audio?

Is the obsolete CD format really that important anymore?

Obsolete means "no longer produced or used, out of date". The vast majority of music content is released and in 16/44.1...today....but it's also in a compressed, bit-rate reduced format, either mp3 or AAC. The vast majority of 24 bit content released and sold today is nothing more than up-sampled from 16/44.1, and therefore no better.

I can't recall the last time I listened to a CD on a CD player, they arrive in the post and get ripped that day.
Perhaps the lack of easily available iTunes downloads at 96/24 isn't the record companies fault after all, but out fault for fighting and insisting we get sold an inferior format?

That's nonsense. Music is purchased for the music, not the level of quantization, except for a very tiny splinter market that buys it only for the level of quantization, not realizing or caring they're still buying the original quantization level anyway. Then there's the true hi-res market that actively seeks out real 24 bit content with provenance all the way back through the production chain. That market is so small as to be static in the statistics.

Why are we demanding mediocrity in audio?

You're being ridiculous now. Nobody does that. They're buying what's available. And frankly, what's available in the most popular music today could be resampled to 8 bits without noticeable degradation. And the rest of it is represented perfectly in 16 bits.

Do we turn down 100W amps because we may only want 2W to use?

No, but some people buy low power amps because the think the active devices and topology used is somehow better. 5W tube amps are actually fairly popular within the tube amp category.

Is that speaker too good for us so we get a lower grade one? Is that steel beam holding up the house too good for the job and we demand a smaller one? Format wars appear to be the only branch of HiFi where people demand less, rather than more.

Nobody demands lower quality unless it's cheaper. And then, cheaper almost always wins the toss. Welcome to the world market. The better more expensive mouse trap doesn't sell, the cheaper ones that still do the job do. Don't be mislead into thinking that the audiophile market is a perfect model of the market in general.

 

[Cutestudio]

Where is 24/96 used outside of audio?

Lots of places, even movie sound tracks have better formats now.
https://en.wikipedia.org/wiki/DVD-Video

Obsolete means "no longer produced or used, out of date".

Yes. 30 years out of date.

The vast majority of music content is released and in 16/44.1...today....but it's also in a compressed, bit-rate reduced format, either mp3 or AAC. The vast majority of 24 bit content released and sold today is nothing more than up-sampled from 16/44.1, and therefore no better.

Sad isn't it.

That's nonsense. Music is purchased for the music, not the level of quantization, except for a very tiny splinter market that buys it only for the level of quantization, not realizing or caring they're still buying the original quantization level anyway. Then there's the true hi-res market that actively seeks out real 24 bit content with provenance all the way back through the production chain. That market is so small as to be static in the statistics.

You're being ridiculous now. Nobody does that. They're buying what's available.

No, not nonsense or being ridiculous, my opinions and view are mine and you are free to disagree, but please don't belittle people you disagree with.
You also appear to contradict yourself here - saying there is no demand for a decent format but then saying people buy what is available.

Threads like this however make me realise that for every audiophile who would want a better format there is a crowd of people just waiting to rush up to explain why they don't need it, and in this way we stay with these pointless discussions and no progress is made. Any time a music industry product manager looks at these threads they see the vocal defence of 16/44.1 which is frankly rather sad.

It looks like for the best formats we can only look to the video industry and hope one day that the format of dedicated audio will ever be as good, but if this thread is any indicator, I think we'll always be using the inferior, obsolete 16/44.1 format and playing with dithers (not a good idea if you are feeding the output into a digital room EQ or crossovers) and upsampling.

 

[Niouke]

I'm not sure why you are saying I'm not looking at the waveform, do you suggest I get inside the wires and watch the electrons drifting back and forth?
Why exactly are you against 24bit, do you need a bigger disk or is it metering charges on the internet?

Here's a study that found 24bit was more perfect than 16bit BTW:

24bits implications:
Bigger storage space
Metered internet charge/Bandwith limitation
Bigger decoding power required -> Battery usage, power demands

I don't think anyone here is "against" 24bits, it is just as you stated "more perfect" and unneeded to achieve the best listening experience possible (for humans at least). Also the marketing hype surrounding it and the attempt to sell unneeded overpriced equipment is disgusting.

 

[Niouke]

regarding DVD video, it has been discussed around here. I'll let the wikipedia page speak for itself:

• AC-3: 48 kHz sampling rate, 1 to 5.1 (6) channels, up to 448 kbit/s;
• DTS: 48 kHz or 96 kHz sampling rate; channel layouts = 2.0, 2.1, 5.0, 5.1, 6.1; bitrates for 2.0 and 2.1 = 377.25 and 503.25 kbit/s, bitrates for 5.x and 6.1 = 754.5 and 1509.75 kbit/s;[12]
• MP2: 48 kHz sampling rate, 1 to 7.1 channels, up to 912 kbit/s.

these are lossy codecs, note that the bitrates given are the total bitrates, all channels added, and are nowhere near PCM 44.1/16

 

[bigshot]

I suppose I could wrap my sandwich in a 40 gallon lawn and leaf trash bag and seal my letters with 3 inch fiber reinforced duct tape, but a sandwich bag and licking the envelope work just as well. File size does matter. I have a large music library with over a year and a half worth of music on it. It all fits on a 2TB hard drive with room to spare. That makes it easy for me to back up, stream over my home wifi network, and load onto my mobile devices. I could store everything at 24/192 but it would be much less convenient and wouldn't sound a bit better. Super high bitrates in blu-ray movies are just as much of an advertising gimmick as "HD Audio" in music. There are people willing to pay extra for sound they can't hear. The nice big numbers give them peace of mind, so the format allows for that. Regular plain vanilla CD sound quality is already into the range of overkill. Lossy audio can be audibly transparent and sound just as good to human ears.