WOW recording in 96 KHz sucks

  • Thread starter Thread starter chadsxe
  • Start date Start date
Bear,

> if someone can't get their shit to sound good at 24/44.1, then going to 24/96 is not going to help them sound any better! <

Amen!

Also, 24/96 requires three times the throughput of 16/44 - twice for the double sample rate, and once more for having 50 percent more bits.

BTW, a few days ago a friend asked my opinion of mixing on headphones so I sent him to your excellent article.

--Ethan
 
i just don't get it.

i thought the only point was to pick up higher frequencies.
once you hit 48 then you are already recording over what anyone can hear, what your mic will pick up, and what your speakers will play.

or maybe i just hate 96 because my computer can't handle more than 5 tracks right now.
 
Matheon said:
i just don't get it.

i thought the only point was to pick up higher frequencies.
once you hit 48 then you are already recording over what anyone can hear, what your mic will pick up, and what your speakers will play.

or maybe i just hate 96 because my computer can't handle more than 5 tracks right now.
Click to expand...

The sampling rate has nothing to do with frequencies.
 
Matheon said:
i just don't get it.

i thought the only point was to pick up higher frequencies.
once you hit 48 then you are already recording over what anyone can hear, what your mic will pick up, and what your speakers will play.

or maybe i just hate 96 because my computer can't handle more than 5 tracks right now.
Click to expand...
There has been some controversey for as long as higher sample rates have existed as to just how much stuff above 22kHz the human ear can actually perceive. There are plenty of studies that show that the human ear can only hear tones up to a certain point (somewhere around 18kHz +/-3kHz or so), yet there are different studies that show that many people can "perceive a difference" between identical recordings of more than just simple tones done at sample rates above 44.1kHz (with a top theoretical top frequency of 22.05kHz).

One reason could very well be that not only does a higher sample rate open the door to higher frequencies, but it also provides finer resolution reproduction of waveforms at lower frequencies. Remember that sample rates is nothing more than the number of times per second that a digital snapshop is taken of an analog value. Even a 10kHz sine wave will be closer to the analog original at 96kHz than at 48kHz. It's not just frequency response that's affected by sample rates, it's also accuracy of waveform reproduction even of lower frequency shapes.

G.
 
Besides trying to record or mix at higher than 44.1KHz sample rates my Benchmark DAC-1 automatically resamples to a higher rate, some plugins do it also like Voxengo Elephant (selectable to 4x sampling rate) and PSP MasterQ to name a couple...from the PSP MasterQ manual:

Frequency Authentication Technique (FAT) Algorithm

The PSP MasterQ, uses our proprietary Frequency Authentication Technique algorithm. FAT is a double sampling technique which allows for proper filter operation at the highest frequencies. FAT utilizes high order linear filters to prevent phase errors and linear errors from the audio bandwidth caused by double order sampling. This technique results in double sampling without sample rate conversion artifacts.

Most Infinite Impulse Response (IIR) filters typically used in EQs tend to focus on phase and linear digital signal errors in the higher octaves. Our FAT algorithm adds an octave above the Nyquist frequency (the Nyquist frequency is measured by sample rate / 2) and shifts phase and linear errors to that frequency region. This frequency region is then truncated just before the output section of the plug-in, meaning the phase and linear errors are removed from the signal. In other words, FAT gives sonic results approaching that of an analog equalizer sampled rather then a typical digital set of filters.
Click to expand...
High sampling rates aren't just for dogs and aliens anymore! :D
 
Ideally, higher sampling rates allow you to set the anti-aliasing filter well above the audible range. All sorts of nasty stuff happens around the cutoff frequency of the anti-aliasing filter, so moving this farther away from the audible range can create noticeable improvements in what IS audible. In practice though, I'm not sure how many A/D convertors with selectable sample rates actually implement a different filter for each sample rate.
 
reshp1 is right.

Although 44.1 is capturing the sound up to 22KHz there is the cut off point where it starts to become inaccurate starts at about 11Khz. The only way to get around these really nasty frequencies above 11Khz were to apply antialiasing filters for them. Thats one big reason why digital sounded so horrible in the beginning. Thats also another reason why there is a big difference in nicer converters. Where it really shines through is in those higher frequencies and the quality the anti aliasing filters the converters have.

I believe the inaccuracies are like nodes. The same issues are also in the really low frequencies below 35Hz. Which is another reason why alot of people will cut everything below that. Its not necesarily a big notice down there but it creates digital nodes on the way up. Try it and you might hear a huge difference in your percieved lowend. Either way thats all energy you probably dont need anyway for most things.

When you sample at higher sample rates the cut off point for those nodes are much higher. 96Khz sampling is way above it and needs much less antialiasing for the frequencies above 11Khz. I believe this is what i believe the difference in perceived highs in. Not that fact that is capturing above 22Khz, its just the frequencies above the cutoff point around 11Khz are much more tighter and accurate.

So in a way with digital we are having to cut corners. Over the years all these little things in the Nyquist theory have been getting better and better with better antialiasing and better sampling but in some ways its still cutting corners. I believe all this stuff i have said is in a more continued and more advanced part of the Nyquist theory if you look it up. The link above is to the basic principles of it.

So is it better to record at 96Khz in the start and down sample? It just depends if you think the antialiasing on your down sampling plugin is better than the anti aliasing on your converter. Either way its gotta go down the 44.1 and do that process. Its really up to you on how it is done. Same goes for dithering from 24 bit to 16 bit. If your ditherer does a crappy job putting in 16 bit noise than it is probably better to keep it 16 bit from the beginning in my opinion.

Ive got a splitting head ache right now so please excuse me if i have made any stupid mistakes in this post :D. But when reshp commented on that i had to throw in my support for it.

Danny
 
Ethan Winer said:
BTW, a few days ago a friend asked my opinion of mixing on headphones so I sent him to your excellent article.
Click to expand...

Where do I find this article? I was planning on mixing in headphones and very nice klipsch computer speakers (a combination of the two).

Also, thanks for everyone who has shared here. It is helping (and confusing me a bit) me a huge amount as I have had this debate personally.
 
a couple of weeks ago i had one of the cd's i mixed mastered by one of europe's biggest studio's, galaxy belgium. They mastering engineer complimenten me about the used converters. And guess what, the recordings took place over a period of 3,5 years, in which we replaced the recording console three times, and the converters 2 times. part of the recordings started at 16/44.1 , the others ended up in 24/44.1. Basic conclusion: always try to keep the recording chain as short as possible with the best equipment you can lay your hands on while recording/mixing.
 
Matheon said:
i just don't get it.

i thought the only point was to pick up higher frequencies.
once you hit 48 then you are already recording over what anyone can hear, what your mic will pick up, and what your speakers will play.

or maybe i just hate 96 because my computer can't handle more than 5 tracks right now.
Click to expand...

Wouldn't a higher sampling rate also help pick up the attack of instruments more accurately, especially drums? I could also see the decay of cymbals sounding more natural. The sampling rate probably affects a lot of other subtle things besides frequency.
 
Guys,

44.1 doesn't give you 22.05khz max frequency captured.

Steep filters need to be applied to get the signal down after the highest desired frequency. They start to roll off at 20k, and are down almost 100db at 22khz. Essentially, you get nothing after 20khz. This is called the guard band. The filters can only be so steep, or distortion creeps in. To prevent aliasing errors, the audio needs to be sampled at a high enough rate to reach a point where the signal is completely gone. The filters are as steep as possible to keep sampling rate as low as possible while still capturing full range audio.

In addition, there is a sync signal in there.

So, in a 44.1 situation, you have 20-20khz for audio, 20khz-22khz for the HF rolloff filters, and 22khz-22.05khz for sync, for a total of 22.05 khz needed to convert 20-20k audio to digital.

One of the big advantages of 48k sampling is the extra room available for the guard band, which allows shallower filters at the high end rolloff.
 
guinsu said:
Wouldn't a higher sampling rate also help pick up the attack of instruments more accurately, especially drums? I could also see the decay of cymbals sounding more natural. The sampling rate probably affects a lot of other subtle things besides frequency.
Click to expand...

No, it doesn't. Dan Lavry's site has a whitepaper on this point:

http://www.lavryengineering.com/documents/Sampling_Theory.pdf


easychair's point about filtering is a good one--in theory, 48 kHz should alleviate some mild attenuation of very high frequencies.
 
mshilarious said:
No, it doesn't. Dan Lavry's site has a whitepaper on this point:

http://www.lavryengineering.com/documents/Sampling_Theory.pdf


easychair's point about filtering is a good one--in theory, 48 kHz should alleviate some mild attenuation of very high frequencies.
Click to expand...


Yup. The filters can start at a higher frequency, and be softer.

You might ask why this wasn't done at first, and why the sync is needed. Why try to cram all this stuff into 44.1 if it leads to compromises? Why not just go with 48khz from the start?

44.1 comes from the fact that digital recording used to be done to digital video, and was stored in the area where the picture was. The amount of samples that could be stored in a video frame works out to be 44,100. The sync was needed for playback, which meant that you had 44k left to cram in 20k of audio and the necessary filters. Not an easy task, and especially early on, there were sonic issues that were part of the reason people complained about CD audio.

Sony and Phillips developed the CD based on the technology of the time, and even though technology and storage methods changed, they and the consumers had spent billions on 44.1 CD technology.

So 44.1 has always been a compromise, a holdover from the early days of digital audio.
 
easychair said:
One of the big advantages of 48k sampling is the extra room available for the guard band, which allows shallower filters at the high end rolloff.
Click to expand...

Right, but do most A/D convertors with selectable sampling rates implement a different anti-aliasing filter for each sampling rate? My understanding was most used the same filter that would work for the lowest common denominator 44.1kHz.
 
reshp1 said:
Right, but do most A/D convertors with selectable sampling rates implement a different anti-aliasing filter for each sampling rate? My understanding was most used the same filter that would work for the lowest common denominator 44.1kHz.
Click to expand...

I don't know, I have reached the limits of my dork knowledge. :D

It would seem to me that using the same high-pass filter for all sampling rates would be the most cost-effective way to go, but I don't know if there is any flexibility in the 48 and 96k standards regarding HPF filter slopes. It may be that the slopes have been chosen and must be used with a specific sample rate, I don't know.
 
mshilarious said:
No, it doesn't. Dan Lavry's site has a whitepaper on this point:http://www.lavryengineering.com/documents/Sampling_Theory.pdf easychair's point about filtering is a good one--in theory, 48 kHz should alleviate some mild attenuation of very high frequencies.
Click to expand...

yet he consistently markets his "pro" products for 96Khz...
http://www.lavryengineering.com/productspage_pro_da924.html

Not to knock Dan's work since he is one of the leading experts, but there are defnitely aliasing artifacts present when you sample near the nyquist limit and moving these artifacts out of the audible range clearly impacts the quality of the sound. artist interpretation or not, most mastering engineers and even Dan himself think 96Khz is a good idea as opposed to simply using 44.1 in anything other than the final product version.
 
gullfo said:
yet he consistently markets his "pro" products for 96Khz...
http://www.lavryengineering.com/productspage_pro_da924.html

Not to knock Dan's work since he is one of the leading experts, but there are defnitely aliasing artifacts present when you sample near the nyquist limit and moving these artifacts out of the audible range clearly impacts the quality of the sound. artist interpretation or not, most mastering engineers and even Dan himself think 96Khz is a good idea as opposed to simply using 44.1 in anything other than the final product version.
Click to expand...

Yes I agree. Dan's argument is against 192, not 96. I cited his paper to rebut the theory that there is something different about attack or decay that is not ultrasonic but can only be captured by higher sample rates. That is not true.
 
You must log in or register to reply here.
Back
Top