Concern over 24bit/96kHz recording

[StarTrek1701]

I am about to begin archiving my Quad recordings, especially vinyl, on my computer. Everybody seems to agree that one should record using 24 bit/96 kHz resolution to get the best result. In fact, I am considering buying a Delta 1010LT soundcard to accomodate such recording. But, what about the sound.

One of my pet peeves about classical re-issues on CD was that a lot of the inherent warmth of those recordings was lost when transferred. Often, they sounded bright (sometimes overly bright), dry, and even hard-edged. Now, I am thinking a lot of that was probably due to engineers "tinkering" with the digital re-mastering in order to take advantage of the higher resolution and dynamic range that the CD provided.

Still, I am worried that recording at such a resolution may bring about the same result. It may not mean a lot with rock or pop material (I just don't know), but with an orchestral peformance, it can mean disaster; resulting in the process not being worth the time or the expense.

Any thoughts or experiences. Thanks.

 

[Almen]

If you were to play the vinyl and listen to it, then switching between "pure vinyl" and adding a 96 kHz/24 bit A/D/A loop, you wouldn't hear any difference. The same with 44.1 kHz/16 bit.

But the best is to record in 24 bits, in case you would want to do some "tinkering".

 

[scifi]

I can hear the difference between 44.1kHz/16bit AND 96kHz/24bit AND LP records.

 

[JonUrban]

Always record your stuff at the highest bit rate and frequency your hardware allows. Save these direct recordings. Put them away on a portable HD or spare HD. You can then "convert" copies of these recordings.

What happens is this. You record the stuff, then "convert" it. You trash the originals because now you have your conversion. However, in the following years, you get better software, better skills, better end-formats to create, and you wish you still had your original hirez files to do over.

In former years, HD space wasn't cheap. These days, you can get a 2Tb drive for around $120. It's a no brainer. Save the originals at 24/96 or 24/192. Then from there, make your working copies at whatever bit rate you like, but always save the masters.

 

[Old Quad Guy]

Follow what Jon has mentioned. I find if anything that higher bit rates capture the "warmth" of LPs because they are able to capture much more information. CD's because of what their bit rates are and people not doing more or less straight transfers from original sources keep fiddling around with the CD files in a futile attempt to capture that warmth or get more from them than what higher bit rates can.

 

[neil wilkes]

Digital & Analogue recording work in different ways. Lots of people forget this.
To try & get a better understanding please consider the following analogy (sadly not original and lifted with thanks from "Mixing With Your Mind" by Michael Paul Stavrou (www.mixingwithyourmind.com) and available from the website. ISBN 0-646-42875-6).
Let us compare recording dynamics to photographing a skyscraper - due to the vast height of the building (range of an orchestra), it's difficult to capturethe entire building in perfect focus, and the 2 formats use different focal areas.
In the case of Analogue, it's as though the focus of our camera is adjusted to halfway up the skyscraper - this puts the middle of the building (where most of the action is) in sharp focus. The areas near to street level (the foyer & food court, for example - analogous to the noise floor) are blurred and indistinct, while the tip of the building (the Radio Tower & Penthouse Suite - the tips of our transients) are also out of focus & somewhat fuzzy. This is acceptable to a degree because the lower floors house mostly tape hiss while the tower only contains fleeting transients.
The Skyscraper is analogous to the volume level of the sounds we record. Analogue captures the most important elements of the sound with pin-sharp accuracy (the middle of the skyscraper where thecritical work is being carried out) while the top & bottom where the activity is less critical is captured less sharply.

A Digital view of the skyscraper is somewhat different. With Digital, it's as if our camera focus has been permanently set to focus on the top section of the building. The communications tower & penthouse are pin sharp, while the remainder of the building slowly goes out of focus from the top down. Why? The pinnacle of the building is the only portion captured or defined by all of the recording significant bits - thus enjoying the sharpest focus in the recording.

If this analogy holds true, wht happens in Mastering (usually, that is) - they remove all the best bits to make it louder!!
So, unlike analogue, where the worst elements are removed, here in the digital world the best.most focussed elements are removed leaving us with a grainier version of the lower level components. We didn't notice the grain before because they were quiet - now they are 10dB louder...........

So to move back into the original point, we record digitally at the highest possible resolution. Even when we are intending to release on 16-bit 44.1kHz CD, we record at the highest resolution. Why? This is all down to DAC and specifically Filter Design - our old, old acquaintances again.
Fact is that in out throwaway world, all DAC is on a tiny chip controlled by firmware, and the filtering is poor (read "cheap") as it seems that it is easier to make a DAC sound good at 96k than it is at 48k or lower. The commonest problems are aliasing caused by too steep a slope on the filters - however research has shown that adding a second gentle slope filter will eliminate all the pre-echo & ringing issues caused by aliasing.
This holds true even if we upsample.
It further seems that the reason for going high sample rate is nothing to do with extended frequency responses, and if you can tell the difference between 48 and96kHz then your DAC filters are not what they should be as there is quite simply nothing up there that anybody can hear past 23kHz at the very highest. If there were, SACD would be unlistenable due to the masses of ultrasonic noise above 23kHz where the noise shaping required to make a 1-bit system work has been shifted into the ultrasonic range.
Oversampling DAC also help considerably, as they first go up a magnitude, thus placing all the noise in a much wider bandwidth, and then going down again to promptly filter most of it right back out again.

Record at 96kHz (192 is overkill and possibly counter-productive too. I will post the research later on) and carry out any processing at the same rate.
Try to avoid downsampling as this can cause artifacts.
Avoid clipping the signal, or trimming the transients to get volume. If you need it louder, use your wrist & crank the volume knob.

 

[kap'n krunch]

Thanks , Mr. Wilkes (not being sarcastic, just respectful) for that great quote and analogy, it is so very true, Cheers...

 

[alk3997]

...

It further seems that the reason for going high sample rate is nothing to do with extended frequency responses, and if you can tell the difference between 48 and96kHz then your DAC filters are not what they should be as there is quite simply nothing up there that anybody can hear past 23kHz at the very highest. If there were, SACD would be unlistenable due to the masses of ultrasonic noise above 23kHz where the noise shaping required to make a 1-bit system work has been shifted into the ultrasonic range.
Oversampling DAC also help considerably, as they first go up a magnitude, thus placing all the noise in a much wider bandwidth, and then going down again to promptly filter most of it right back out again.

Record at 96kHz (192 is overkill and possibly counter-productive too. I will post the research later on) and carry out any processing at the same rate.
Try to avoid downsampling as this can cause artifacts.
Avoid clipping the signal, or trimming the transients to get volume. If you need it louder, use your wrist & crank the volume knob.

Neil, I am curious about a couple of things you said and comparing against my own experience.

It started years ago when I looked at a few CD mastering jobs and realized that only 14-bits of the 16-bits were being used overall. This got me to thinking about mid-bass response. When mid-bass instrumentation is soloed, many mid-bass instruments show up with about 6-bits of dynamic range in a CD recording. In order to balance that with the rest of the band, those 6-bits are often towards the lower six bits of each 16-bit word. Listening to a 6-bit recording is not an enjoyable experience but I'm not sure that is valid in this case since the overall recording is still 16-bits.

My question is whether this is why the warmth of those particular frequencies is lost when recording to CD-specs? 20-bit and 24-bit recordings seem "warmer" but I've wondered if this is only because I know they are 20-bit and 24-bit recordings.

I know the inventor of the CD format (I forget his name) stated that 16-bits was the minimum for a good recording (I think he originally wanted to go with 14-bits) but I'm not sure he ever thought about people only using 14-bits and leaving out the 2 most significant bits while mastering.

Andy

 

[kap'n krunch]

Neil , how about using dithering when mixing down? I usually do, but ever so slightly(light settings)...

 

[neil wilkes]

Good questions - this is turning into a fun thread.
The biggest problem with comparing vinyl to CD and digital in general is that you're usually trying to compare Apples with Oranges.
The systems are completely different, and work in a different manner which was the point of the above analogy.
"Warmth" is such a subjective term, and what we think of as "warmth" can come from a variety of places - the transformer in your amplifier,
any tubes that might be present, even discrete tone controls set flat can alter the sound of audio when played through them.
The Pultec EQP-1A equalizer is a classic example of this. It's a well known trick in the studio to use one of these in the signal path
with everything set to off - just leave the audio passing through it with no cuts or boosts applied.
The audio always sounds better (well, nearly always) - even when you factor in and adjust for the 1.3dB boost it also applies.
It literally makes things sound, well, warmer.

The best reason for utilizing a 24/96 resolution is that this will be about as good as it gets. You are capturing the audio at the best resolution you can.
(I have read - but cannot locate & must look harder - documents saying that going over 96kHz is counter productive for some reason although I cannot remember what)
As long as this original transfer is saved (and ideally, tracked with a new stylus from a properly weighted tone arm that has been centrifugally balanced too,
going through a high quality PreAmp with the correct curve applied for the year & label the record was pressed if we really want to get jiggy with it)
then all edits can be made to copies, or done in a non-destructive environment so the originals are always preserved.
Once we get the beast into our digital system there are some really cool ways to get warmth back in.

@Kap'n.
Depends on what we are going from and to, really.
If I am rendering for CD I always dither as the very last process I do. Often flat triangular, not necessarily noise shaped either.
If Noise shaping is to be used, I will usually go with the Apogee UV22 system, set to autoblack & low noise level.
There are other considerations though, depending on the DAW being used. If it uses a fixed-point mix engine then every plugin you use will be dithered on the output back to the DAW.
Also, if going from 32-bit float to 24-bit fixed you could argue there is no need to add noise by dithering.

I have also heard of cases where mix engineers actually like to add a bit of very low-level noise to their mixes.
Cannot remember type or amounts off the top of my head though, so will say no more until I check.

 

[DKA]

Also, if going from 32-bit float to 24-bit fixed you could argue there is no need to add noise by dithering.

I was waiting for you to say something like this and am glad to hear this. It confirms our decision to leave dithering out of the upmixing progress when working with Plogue, as Plogue does all its work at 32-bit float before I move it down to 24 bits before encoding to DTS.

 

[ArnoldLayne]

Interesting thread indeed.

As a young guitarman I was taught that the "warm" sound of tube amps and the "coldness" of transistor ditos was because of the way they emphasis odd or even harmonies. Could something similar happen in digital sampling, e.g. making the sound "colder"? Iff so, can it be avoided or filtered away?

 

[alk3997]

I am about to begin archiving my Quad recordings, especially vinyl, on my computer. Everybody seems to agree that one should record using 24 bit/96 kHz resolution to get the best result. In fact, I am considering buying a Delta 1010LT soundcard to accomodate such recording. But, what about the sound.

One of my pet peeves about classical re-issues on CD was that a lot of the inherent warmth of those recordings was lost when transferred. Often, they sounded bright (sometimes overly bright), dry, and even hard-edged. Now, I am thinking a lot of that was probably due to engineers "tinkering" with the digital re-mastering in order to take advantage of the higher resolution and dynamic range that the CD provided.

Still, I am worried that recording at such a resolution may bring about the same result. It may not mean a lot with rock or pop material (I just don't know), but with an orchestral peformance, it can mean disaster; resulting in the process not being worth the time or the expense.

Any thoughts or experiences. Thanks.

Before I forget (again), there is a trick to recording 5.1-channel with the 1010LT that I wanted to pass along. If your recording starts out fine and then x minutes into it you start hearing "fuzzy" distorted sounds where there shouldn't be fuzzy distorted sounds, then check your computer's BIOS settings. Disabling SpeedStep (also known as C1E) will solve this problem.

It seems the act of slowing down the clock will throw off the 1010LT's timing. This might apply to 4.0 recording, as well.

Andy

 

[geminigod]

if you can tell the difference between 48 and96kHz then your DAC filters are not what they should be as there is quite simply nothing up there that anybody can hear past 23kHz at the very highest.

Be careful not to confuse sampling rate with sound frequency. They are measured with the same units, Hz, but are two totally different things. 44.1, or 48, or 96 kHz is talking about sampling rate. 22-23kHz is the frequency of sound at which the human ear is no longer able to hear (and you wouldn't want to hear anyway, unless you are trying to make music for your pet gerbil.)

 

[sukothai]

Be careful not to confuse sampling rate with sound frequency.

I think Neil was referring to the Nyquist–Shannon theorem. A sampling rate two times the max audible 23kHz is less than 48kHz; so there is no need for a higher sampling rate.

I've heard an argument you need a sampling rate of 60-70kHz to preserve ultrasonic harmonics; which have a pleasing effect on the brain. However, 48kHz works for me.

 

[geminigod]

I think Neil was referring to the Nyquist–Shannon theorem.

Yes, you are right. I realized after I posted that that that (could I possibly string any more "that's together?") was what he was referencing, but just not by name. Some thing Neil had said earlier in that post threw me off a little when he made the argument in favor of 96khz recording, which generally is advice that runs counter to the Nyquist nazis. The post was a little wordy and I just didn't read carefully enough. The reason Neil suggests 96k is preferrable to 44.1 or 48 is because cheap DAC's do a better job processing it. He probably has a very valid point there. The strict interpretation of the nyquist-shannon theorem is to say that there is no difference in quality between 48k and 96k, except perhaps to your dog or pet gerbil chatting away at ultrasonic frequencies. The only slightly plausible argument against this is what you mention in your 2nd paragraph, Sukothai, which is some sort of Ultrasonic argument where the sounds we can't hear theoretically still have an impact on the audible sounds we do hear or on our general "feeling" about the music.

While that sounds good on paper, I am inclined to agree with what Neil stated in his post upon re-reading. Theoretical science is like dating a girl. It is new, exciting, and relatively straightforward. Applied science is marrying that girl and trying to make the relationship work for 30 years. Applied science is messy and the pretty math is distorted by an enormous number of uncontrolled variables. Given the power and storage capacity of modern computers, it can't hurt to record an initial pass in 96kHz, even though it will ultimately be down-sampled, and even though there are a number of other factors that will unquestionably contribute more to the quality of sound than increasing your sampling rate above 48kHz.

Or can 96k actually hurt the final product? One topic which I would love to hear more feedback about is the effectiveness of various software and filtering tools at different sampling rates. This is probably a difficult topic to research or discuss without getting software specific, but I have seen the argument made that some editing tools are designed around a 44.1kHz sampling rate and the 96kHz sampling rate can actually become problematic and result in too much sound data being cut out. I have occasionally experienced problems personally using some noise filters where too much sound is occasionally cut out for some weird reason on a 96k recording, but I have never attempted a side-by-side comparison of the same material at two different sampling rates.