so if there is no difference in the subjective quaility then its pointless to record and replay at 24/96?

Playing at anything higher than 16/44.1 or 16/48 is pointless, but recording is not. If I recall correctly, using higher sampling rates reduces aliasing that can occur during equalization or DSP during mastering. It would be analagous to how one edits an image at twice the target resolution, then resizes to half to eliminate any aliasing.

At recording stage you want the maximum quality possible. So you can "mess around" with the sound using digital and analog, destructive and non destructive processes until you get what you want. Because you start with more resolution than necessary you can afford to lose some of this resolution before reducing the output to 16/44.

So it is a waste of time for listening but not for recording.

This depends on what one is recording and what will be done with the recording. If the source is low noise enough to provide the dynamic range, 24 bit is definitely better. If the source is something like an LP or a cassette, the background noise is too high for it to make any difference.

If you are going to do much processing of the recording, such as noise reduction, EQ, reverb, mixing with 20 other tracks, ... 24 bit is better (floating point, 32 or 64 bit is better than 24 bit). However, for noisy sources, as mentioned above, 24 bit or floating point might make a difference if there is to be a great deal of this post recording processing, but most of the time you can’t do enough to make any difference you can either hear or measure.

It is easy to show aliasing images when recording -- test tones -- at 44.1kHz, if those tones go above 22,050Hz. Unless the source frequency is higher than the Nyquist limit for 88.2/96kHz (which, I think, does not exist in music) there will be none of this aliasing when recording at the higher sample rate. By recording frequencies higher than 22,050Hz at a higher sample rate (e.g. 88.2kHz or 96kHz) you can eliminate essentially all of this in the final product resampled to 44.1kHz.

Some people claim to be able to hear this aliasing result from 44.1kHz recording of some music, but this is all subjective reporting. Actual testing has indicated otherwise; in blind ABX testing people are unable to differentiate between the different sample rates. Thus there is no practical benefit from recording most real music at greater than 44.1kHz.

A few instruments do produce a significant amounts of energy at higher frequencies than can be captured at 44.1kHz, and one published report claimed to measure responses in listeners which the researchers believed were due to these higher frequencies. None of the listeners were aware that they were being exposed to the higher frequencies, however. Thus even when recording such instruments live (assuming you are using extraordinary microphones that can capture them) there will probably be no noticeable difference in the final result. With any other music there most certainly will not be.

When people here talk about resampling at a lower rate is the use of an anti-aliasing filter assumed?

To me, Emon's statement about "how one edits an image at twice the target resolution, then resizes to half to eliminate any aliasing" is either nonsensical or misleading.

EDIT: I didn't realize Emon was talking about video earlier, but still don't think it adequately addresses the concept of aliasing as it applies to audio.

I agree with you, greynol.
When aliasing occurs in audio it usually generates false low frequencies. Therefore, downsampling won't remove aliasing products once they're introduced.
That's why one has to fight the problem before it occurs using anti-aliasing filter not just for resampling but also when doing analog-to-digital conversion.

Non-linear DSP operations will cause aliasing, so 96khz is of huge benefit for mixing.
Generally speaking, an equation of polynomial degree n requires n times oversampling. This is usually accomplished by upsampling, applying the DSP, and then downsampling back. Since there's rarely enough CPU power to do steep filtering on downsampling, a bit of aliasing creeps back in. Using 96khz allows all this aliasing to remain out of hearing range, and ultimately disappear when the track is finalized to 16/44.1


Huh?!?! This makes no sense unless you have SEVERE aliasing

That's right, I should have drank my morning coffee before I wrote that!
Sorry…

Still, a 40kHz tone present in a 96kHz sampled recording doesn't simply disappear when down-sampled to 44.1 kHz. It's image will fall squarely in the audible range.

How so? The filter applied before downsampling should take care of it.

In extreme cases, music might reach 30kHz, say even 35kHz. This is well within the bandwidth for 88.2kHz recording, so there will be no aliasing. Resampling such a recording to 44.1kHz with good software will not result in aliasing in a lower sample rate product.

Where the input frequency runs above the Nyquist limit, the alias image in the recording is almost at full signal strength for the first few kHz below the Nyquist limit. If the input has a high signal level (unlike any music) that image is very noticeable (At least in measurements. Can you hear anything at that frequency?)

The image can also be traced, at slowly decreasing levels, down through as much of the audio band as there is something above it to produce the image. Thus it is possible to record at 96kHz and end up with an very faint alias image in the 44.1kHz resampling if the input has a high level and extends far enough above 96kHz, but we are not talking about music anymore.

With any real music, and decent software, recording at a higher sample rate will insure no aliasing image if you want a 44.1kHz result. However, as I stated in my first post, it is extremely unlikely you will get an image that is detectable, recording at 44.1kHz, when your input is music rather than test tones.

If my post isn't clear enough, I am saying this is not true. Good software (e.g. CoolEdit), properly used, will not produce any aliasing image in the lower sample rate product. This is easy to demonstrate.

Hence my question earlier...
Judging from the responses, I'll assume the answer is yes.

Even without audio outside the target sample rate, the filters are highly recommended. I don’t recall the details, but some in some experiments I did some time ago, filter vs no filter was quite audible, and all my votes for quality went to the filter use.

Further mulling leads to more unsureness about those experiments I mentioned. Some off-normal settings gave me results where I could hear a detrimental difference, but I’m not sure it was due to the anti-alias filters. Some kind of variable filters are also used in the Quality setting, according to the CoolEdit help files, so it may have been through playing with that. It is easy to see the differences with either setting when using test tones.

so it seems that everyone is of the same opinion, that replaying audio at 24/96 is a complete waste of time and there is no audible difference, so it must be just marketing from people like chesky, who sell 24/96 dvd's, would this be the case?

Many if not most audio professionals record in hi-res formats (like 24/96). IMO it's reasonable to offer a lossless version of the master to the public.
The 16/44.1 version might sound identical, but most likely not better than the hi-res version (unless your playback chain is broken).
Does Chesky offer all releases as cd as well ? If they don't then the dvd ($24.98) seems a bit overpriced compared to the cd ($15.98) and sacd ($19.98).
If you prefer, you can always convert the 24/96 version to 16/44.1 yourself, but that would be a real waste of time IMHO

You should be able to search out the thread, from the past few weeks, that covered the recently published report on the blind ABX tests with hundreds of people over more than a year’s time. Using SACD and DVD-A disk as one element and the same resampled to 16/44.1kHz as the other, noone could tell which was which.

Chesky is a exception to the general rule, as they try for the highest quality in all their products, but when the same music is offered on both CD and higher resolution disks, the mastering is often markedly better on the high resolution version. Thus if you compare those with the CD version, you will find superior audio quality. It just has nothing to do with the higher resolution.

Ahh, thanks.

I might add that analog brick wall analog filter is not very practical because in needs infinite time to operate. To make the time finite causes phase distortion in the signal depending how deep the filtering is.

On DAC you can always overcome the analog filter restrictions by resampling to much higher frequencies and push the analog filter way out from audible range. I find this kind of DACs really good sounding.

On ADC you really want to input highest possible sampling frequency to have more loose requirements of the analog filters to push any anti-aliasing effects way out from audible range. In the digital domain you can resample to 44.1kHz which is adequate.

i see that explains alot- in the mastering stage is it better because it stays at 24/96?
but at the end of the chain are you not relying on the quaility of the DVD player's
convertors, so if there terrible (are the all the same?) you lose all that resolution anyway?

Its not really better, its just easier to work with if you have some headroom.



That resolution never does anything anyway, so the quality at the end isn't the issue. The limit is generally biology.

Trying to answer the question ..

It's not a waste of time, as it doesn't cost any extra time, depending on your opinion it could be a waste of space though.

I think with 24bit there is always the benefit of not having to use dither.

Dither is not anything to get all concerned about when it's done as a final step when going to CDDA, as has been said many many times before. Furthermore, the quality of analog reconstruction from CDDA is far from being night and day when compared to something sampled at 96k as some seem to insinuate.

AndyH-ha,

Your discussion of sampling at a higher rate and then downsampling to 44.1kHz is true in the practical sense - but of course in the theoretical sense it is meaningless. Almost all good converters oversample by many times. The fact the converter uses a poorer filter to get down to 44.1kHz than Cool Edit, SRC, or whatever is only a "practical" detail - in theory the oversampled converter can output 44.1kHz with as little aliasing as you desire.

There is the same benefit to playing CDs back through 88.2kHz DACs: you can use better filtering than is typically available off the shelf. As with ADCs, the difference is measurable and audible with test signals, but hasn't been proven with real music.

Btw, you should take a look at this month's Hi-Fi news (UK audiophile mag) which discusses filterless DACs (and ADCs!) and makes claims for their superiority. It was so bad I had to buy it. If it weren't for copyright concerns I would scan the entire article and put it on line for your all to have a good laugh at it.

Cheers,
David.

I don’t pretend to know what is going on, I only have the empirical evidence. Oversampling at 64X and 128X are the specified norms for modern soundcards, yet the alias images are very strong (relative to the input audio) for the first few kHz below the Nyquist limit, and they continue down to a quite low frequency if the higher frequency input is there to be ‘reflected.’

Delta Sigma converters, operating at a much higher sample rate than SACD if I read specs correctly, are at the front end of the processing in almost all modern soundcards. Their output bit stream is decimated down to the target sample rate/bit depth. Brick wall filters are no longer (much?) in use. Just what is the weak link, from the test signal pont of view? My understanding of both the electronics and the theory is too primitive to provide the insight.

I’ve read a few bits about the guy in charge at the LSO. They put out fairly inexpensive CDs of their own stuff, which are reportedly very good. He records at 176.4kHz and resamples in a very conventional way which, he claims, involves no anti-alias filters. The LSO web site , however, makes no mention of this practice, last time I looked.

it seems to me the case is this: if you can record at higher sample rates do, someday empiracal facts will put the whole thing to rest, if your proven wrong so what, go back to 44.1 if your right- great you've made a great recording that will stand the test of time...!
i also think that when mechanical parts are removed from the system as in DVD -CD and data is streamed direct from hard drive to the DA converter all this confusion about rates will be over, we'll all listen at for example 24/32/64 bit 96/or even up to the 300K sample rates if we ever wanted too, i give the cd 5 years max to be extint, soon after the DVD! my 2 cents

I tested a cheap realtek onboard DAC a while back and got no aliasing from 21.5k or so and below @ 48KHz (only native sample rate). 2.5KHz isn't a bad transition band at all, and anyway "very strong" is pretty ridiculous I think. Seemed to loose 20 or 30dB a Khz.

Perhaps you somehow tested an amazing bad DAC?




Mechanical parts have nothing to do with sample rate. And anyway, if you want a solid state system, buy a flash player. No need to wait for that, they've been on the market for years now.

I’ve tested some M-Audio and Echo cards, and a soundblaster that was especially bad at anything other than 48kHz, due probably to its resampling. Having posted on the topic in a number of forums over several years, a number of people have reported back similar results with other cards, including a couple of significantly more expensive ones than those to which I have access .

I haven’t tried to measure that I can recall. I’ve observed the recordings in CoolEdit’s Spectral View (I’ve posted images here in the past). Near the Nyquist limit, and for two to three KHz below, the alias image is quite bright but not as bright as the input signal. By then it is quite clear that it is less energetic and fading, however. One might think it soon gone, depending on the Spectral View settings, but turning the resolution up to 200dB (120dB is default) shows it continuing, at slowly decreasing brightness, down to at least a few kHz from the bottom of the display.

I still have a test on disk where I used tones stepped at every kHz, from 22kHz to 30kHz, rather than the continuous sweep tone up to 48kHz I was talking about before. The image from the 30kHz tone (at 14kHz) is still clearly visible with Spectral View default settings. The first image (a little above 21kHz, from the 23kHz input) measures 18dB down from the input. This is admittedly attenuated relative to the over-frequency input, but still glaringly obvious. With a continuous sweep tone, that part immediately below the Nyquist limit is clearly brighter.

Note, I have maintained that the alias image’s impact on any real world music is likely to be insignificant, as far as anything one hears (some people disagree). I don’t see the aliasing as a practical problem.

I think empirical facts have already put the whole thing to rest, as far as people interested in empirical facts are concerned. There isn’t anything there to justify higher sample rates for music. How, audio reproduction is one of those concerns, such as religion, where the empirical method holds no interest for many people.

I just located a recording of a sweep tone. The image is quite faded by 19.5kHz, but I can see it all the way to below 1kHz, at the default spectral view settings. This means it is definitely still well above the soundcard noise floor (that part of the recording from input below the Nyquist limit measures -13.4dBfs).

Andy,

I wasn't disagreeing with your practical findings - I've done the same experiments and found the same results. I'm not sure what Mike's talking about.

The simple point is that I believe these converters (both ADCs and DACs) often use 128 or 256 tap FIR filters (or sometimes unknown IIR). Cool Edit Pro seems to be using about 4000 tap FIR filters, depending on the setting. If the converters were doing the same, you wouldn't see any benefit from recording at a higher rate and resampling in CEP.

Cheers,
David.

i use apogee mini-me dac, and it definitly sounds a lot better then any dvd player i have used, that was the point i was making, i have listened to the smae material on a off the shelf dvd and then on the apogee, and it is better! so i think one day everyone will have this quality because it will become cheaper.

Testing how bad resamplers sound isn't very interesting, so you need to use the native sample rate.



Ok but how many dB at say, 20KHz? How many at 18KHz? Unless its very close to peak, its irrelevant, since you're not going to hear it. Saying its "bright" doesn't really mean much.

I started out stating that I don't believe anyone is likely to hear it. I never suggested otherwise. Of course, I don't know that all of those who claim to hear a difference are only imagining it, but I suspect this is so.

I simply pointed out that aliasing images are created. Recording at 44.1kHz, when input signal at greater than 22kHz exists, will produce aliasing distortion. What is your point?

The thing with high-res formats is that there are in fact two things that change. While I think it's been made quite clear that there's no *audible* difference between something *played back* at 44KHZ VS. 96 or 192 KHZ (as even 44 goes well beyond the range of human hearing), my quesiton is what about the change from 16 to 24 bits? IIRC, this effects dynamics, both in giving it a deeper noise floor and in allowing the sound more room to swell, giving it a more lifelike, "in the room" feel, for those sound engineers that know how to take advantage of such. Is this accurate? Ignoring samplerate completely, is there still any sort of practical reason to play back a 24-bit source over a 16-bit one? Or is the difference in bitdepth as inaudible as the increased samplerate?

It is too hard to keep all these threads straight. Without re-reading all of this one I can't say if the facts has been discussed here, although they certainly has been many other times. In addition to the advantages of working at a higher bit depth, preferably floating point, while mixing and mastering, it is easy enough to demonstrate an audible difference between 16 bit and 24 bit at very low signal levels -- with test tones. However, for some years now, I've been asking on various forums for any 24 bit format champion to provide a sample of real music where a 24 bit master can be ABXed against a properly resampled 16 bit copy of same. So far no one has been able to find one.

lets start a new thread... this is dead and confused!

I don't think so.

You should read TOS #8, however.

I'm guessing this is why people have stopped responding to you, but I could be wrong.

When I first read this, I briefly wondered about how you tested whatever led to the statement, but mainly I just let it slide by as something unknown to me. Now after a little thought, I don’t understand it at all.

An audio file doesn’t have any frequency information, other than the sample rate in the header. It is simply a collection of sample values, which are nothing more than amplitude measurements. Frequency is derived from the rate of change of amplitude, progressing from sample to sample. That rate of change cannot be greater than that which will produce the maximum frequency the sample rate supports.

It is possible to change the header sample rate, and thus play the samples faster to produce higher frequencies. Changing it from 44.1kHz to 88.2kHz would make a 10 second, 20Hz to 22kHz sweep tone play as 40Hz to 44kHz in 5 seconds. It is not otherwise possible to play a 44.1kHz sample rate file back at 88.2kHz, in any way I understand. Therefore I can’t think of a way your statement about reducing aliasing at playback works.

It is of course possible to play 44.1kHz files through a DAC that will also play 88.2kHz files, but that doesn’t seem relevant to what happens vis a vis aliasing by moving the ADC sample rate up above the basic input signal frequencies. Perhaps this means that you means something I haven’t even guessed at.

I think David was saying that taking a 44.1 kHz file, upsampling it to 88.2 kHz and then converting to analog via a 88.2 kHz DAC plus suitable analog filtering can be superior to the more direct route because it is easier to filter out any components above Nyquist without altering the audible data.

Yes, just that - that you can use a better anti-image filter in software, or as a pre-process, than would be employed by the DAC itself - i.e. exactly what you've been talking about. Nothing more.

Cheers,
David.

[delete post]
ah, never mind, people already explained that David's talking about CDP oversampling...a rather noncontroversial technology at this point, I'd hope.

I've been out of the game for a long time now, but what type of digital filter has a frequency response with no images of its own?

Seems to me reconstruction still needs a low-pass.

@AndyH-ha, to add to what pdq already said, oversampling allows for easier filtering because it pushes the adjacent image farther away giving you the ability to use a filter with a much larger transition band.

Edit: I wanted to add that oversampling is a lossless process.