Audible difference 44.1kS/s vs 96kS/s? (files included)

[dcollins]

Originally Posted by Key

I think I am mixing up or basically don't understand the differences between asynchronous sample rate conversion and a digital to analog to digital conversion with asynchronous clocks on the ADC and DAC.

Synchronous means the input and output clock frequency/phases have to be connected, asynchronous means they aren't.

Is that enough?

DC

[Key]

Probably not for me. But anyway in the one case I am pretty sure there shouldn't be weird ringing like you would get with an in the box SRC and an uneven resample. And I guess if you get the noise floor, imd, THD, PD blah blah good enough couldn't this work better than an itb SRC at least in terms of perceivable artifacts like ringing or echoes? Or is that what I am confusing echoes with rings? what was the question lol.

[WAVERUNNER]

the second one sounds more truer to the original. but the first sounds brighter. For production, I would go with the second.

[dcollins]

Originally Posted by Key

Probably not for me. But anyway in the one case I am pretty sure there shouldn't be weird ringing like you would get with an in the box SRC and an uneven resample. And I guess if you get the noise floor, imd, THD, PD blah blah good enough couldn't this work better than an itb SRC at least in terms of perceivable artifacts like ringing or echoes? Or is that what I am confusing echoes with rings? what was the question lol.

Every filter rings. The box doesn't care.

FIR is the weird one in terms of ringing or pre-echoes.


DC

[UnderTow]

Originally Posted by WAVERUNNER

the second one sounds more truer to the original. but the first sounds brighter. For production, I would go with the second.

I am always amazed at what people type.

Anyway, for some strange reason PCABX or WinABX are not outputting any sound... I can't do a proper listening test. When quickly throwing them in Winamp and playing them once without concentrating too much, nothing obvious jumps out at me...

Alistair

[Audiop]

Originally Posted by ISedlacek

So you recorded 3 clips in 96k, then converted them to 44k and then back to 96k ?

Yes!

Unless I miss something in your description, the real point that shows the difference is to record the same source once in 96k, second time in 44k (using the same recording chain). Then to convert 96k file into 44k (using high quality SRC - like r8Brain Pro) and to compare the differences in 44k a to see whether recording originally in 96k brought any sound difference comparing to recording directly in 44k.

Yes Ivo, that's a test that make sense as well. The reason I didn't such a test this time is because I have allready nailed a small but obvious coloration from the 44.1kS/s loop of the converter I mainly use (Lynx Aurora 8). This time I wanted to get a feeling for the audibility (or absence of) high frequency content in recording of instruments with significant output above 20kHz.

If you recorded only in 96 k (and that too using rather low end quality converters) and then did just some SRC processes, I am afraid it does not say much - maybe something about the isotope SRC algorithm.

Audiofire 4 is an excellent converter and certainly good enough for this purpose.

But yes, obviously this test is about two things basically.. a) wether or not an audible difference is there when removing everything above 22khz and b) the quality of iZotope SRC.

If there is an audible difference we can move on and try to investigate if a or B or possibly a and b are the explanation.


/Peter

[Audiop]

Originally Posted by neilwilkes

It's cheaper to design & build a converter to sound good at 88.2/96 than it is to make it sound equally as good at 48/44.1.
Bob Katz explains it much better than I can in the 2nd edition of his excellent book
Bob Katz - Mastering Audio
Symptoms can include any or all of these:
higher distortion, ripple, ringing (*) and aliasing.
* = this one is very common in poorly designed Linear Phase SRC too

Yes, I'm aware of the things you write and IMO it makes more sense to use a cheaper converter with a software SRC than to buy something at ten times the cost. I prefer to put that money into mic's or a new production room.

As far as actual HF content goes, can you honestly hear above 22kHz?

Nope. The highest I've heard are 21.5kHz, at high levels I should add. At normal levels my hearing rounds off at 17kHz or so.

The question that often is raised if it's possible to hear HF in transient sounds even if we can not hear it with more stationary signals. So far no-one has been able to prove that is the case AFAIK.


/Peter

[Audiop]

Originally Posted by Key

haha no but I just check recently and I can still hear 21kHz.

I downloaded the files but I would actually prefer a whole song to increase the odds of audible artifacts. And it makes the listening a little less like a Chinese water torture test for your ears.

Motivation is good of course but basically it's easier to hear small differences on short clips.

/Peter

[PatrikT]

Originally Posted by Audiop

Yes, I'm aware of the things you write and IMO it makes more sense to use a cheaper converter with a software SRC than to buy something at ten times the cost.

This would imply that the design of the analog side of converters has nothing much to do with the end result, but it surely has.

I have to listen to 44.1 kHz a LOT, maybe others don't.


Regards
Patrik

[Key]

Originally Posted by Audiop

Motivation is good of course but basically it's easier to hear small differences on short clips.

/Peter

Not for me personally. In my experience I need a variety of sounds to really expose weaknesses in an SRC. And if it's a song I have no problem listening to it over and over but those clips you posted I only listened a couple of times before it became very monotonous and I just shut them off. It makes the perceived importance small imo when you are just using 3 very short sounds and not a song.

[Audiop]

Originally Posted by PatrikT

This would imply that the design of the analog side of converters has nothing much to do with the end result, but it surely has.

I have to listen to 44.1 kHz a LOT, maybe others don't.


Regards
Patrik

No why?


It goes without saying that any circuitry that the signal pass thru is important and having modified several CD-players with new clock's and analog circuitry I'm fully aware of the importance of those parts but AFAIK the clock and analog does not change when you switch between rates on the same converter.. ;-)

If a converter is transparent at high rates that means that the clock and analog circuitry is flawless and any difference when going down to 44.1kS/s means bad implementation on the digital filter.


/Peter

[Audiop]

Originally Posted by Key

Not for me personally. In my experience I need a variety of sounds to really expose weaknesses in an SRC. And if it's a song I have no problem listening to it over and over but those clips you posted I only listened a couple of times before it became very monotonous and I just shut them off. It makes the perceived importance small imo when you are just using 3 very short sounds and not a song.

If you test hundreds of devices with hundreds of listeners over thousands hours over a couple of decades chances are you'll find that short clips is a sharper instrument when looking for audible artefacts.

Of course there are exceptions though.

Most people that use songs when testing seem to look for a "hard part" and focus on that anyway.

It also goes without saying that you need a variety of sounds depending on what kind of errors you are looking for in gear but when looking for an answer to the question(s) that led to the test in this thread it's pretty obvious why I choosed the instruments I did. I'll do more tests though and will use cymbals as well then.


/Peter

[Key]

Well I'm not so sure about that exactly but you could be right I guess. I just think this is too close to a tone test or any other monophonic test tone. You need polyphony for certain things to be present like imd. Anyway don't get me wrong I said on another forum that I was pretty sure if you did this type of operation and took out the variable of a 44.1k DAC vs 96k DAC by upsampling the 44.1 back up to 96 that most everyone will not be able to ABX between the 2. I just think your examples aren't "real world" enough to convince everyone. Real music is just much more complex than a monophonic high pitched tone with some overtones.

[PatrikT]

Originally Posted by Audiop

No why?

You could benefit from something that actually perform very well at 44.1 kHz if the context is mastering, which I GUESS is the context of this forum.

96 kHz into and back from the analog - whatever, but in the end of the day you will have to listen to 44.1 kHz in order to hear what you actually provide to the majority of the world. At a professional level.



Regards
Patrik

[Audiop]

Originally Posted by PatrikT

You could benefit from something that actually perform very well at 44.1 kHz if the context is mastering, which I GUESS is the context of this forum.

Sure, but not necessarily mastering at/to 44.1kS/s.

Also I guess a 44.1kHz DAC would be more essential than AD as you can always record at high rates and do SRC in post. I stay ITB and want pristine AD.

96 kHz into and back from the analog - whatever, but in the end of the day you will have to listen to 44.1 kHz in order to hear what you actually provide to the majority of the world. At a professional level.

Yes, but there are a market for higher resolution media as well. Not saying that we/they/anyone really need that from an audible point of view though but I'm interested to find out what's thru and have limited interest in what the marketing department of some companies say.

I want to mention again that the basic purpose of this test is to see if people really can hear a difference on audio limited for redbook and higher resolution material.


/Peter

[Audiop]

Originally Posted by Key

Well I'm not so sure about that exactly but you could be right I guess. I just think this is too close to a tone test or any other monophonic test tone.

There's a big difference between a sine and a transient signal of these types. Shakers are very complex and far from a monophonic tone.

You need polyphony for certain things to be present like imd.

Yes and no. You need something more than a single sine but IMD comes from the same nonlinearity as HD and a wide bandwith transient signal has the potential to produce these distortion products. Also these tests is not about hearing distortion but high frequency content.. I think I have mentioned that once or twice.. ;-)

I just think your examples aren't "real world" enough to convince everyone. Real music is just much more complex than a monophonic high pitched tone with some overtones.

First, these instrument are real instruments and what constitutes music is a personal/cultural/subjective thing. I have never said that these clips are meant to be the end to all other tests, this test (or this type of test) is one important piece of a puzzle IMO.

Also my intention is not to convince but to collect knowledge.


/Peter

[David Spearritt]

So all you chaps who are complaining about Peter's test methodology, obviously didn't hear any difference in the tests right?

Could you confirm?

[Arksun]

Originally Posted by David Spearritt

Huh?

He's probably referring to how a lot of plugins perform and sound better when running at a higher sample rate, particular anything with envelope following like a compressor. They tend to sound less clicky at higher rates. Early soft-synths benefitted greatly from working at a higher rate to reduce nasty aliasing in the higher notes. Today however soft-synths have built in oversampling or other band limiting methods to reduce such aliasing issues at 44/48.

eq's as well, more accurate curve shape at higher frequencies (though again, eq plugins written today can employ built in oversampling or other algorithmic methods to retain shape like the Eiosis AirEQ)

It just sucks that working at 88/96 literally doubles the cpu usage, as a heavy soft-synth user I can never have too much power. D-Cam Synth Squad will bring a core i7 to its knees with ease.

[Key]

Originally Posted by David Spearritt

So all you chaps who are complaining about Peter's test methodology, obviously didn't hear any difference in the tests right?

Could you confirm?

I didn't. But I didn't try very hard because I found the sounds annoying. It seems to be a pretty simple confounding variable if you ask me. Give me some music that is longer than a few seconds. I like John Cage and all but damn I am not going to get into the whole "what makes a song a song?" argument. You know what I mean Audiop lol.

[myrtlebacker]

Originally Posted by Audiop

Thanks for posting. Do you have a link to that software? I have been searching
for ABX for Mac.

Can't find a link to it, I don't remember where I got it from. It says
ABXer Copyright 2009 Empty Music Inc. Can't say if its OK to distribute it or not.

[Luna Sound]

Originally Posted by Audiop

Thanks for posting. Do you have a link to that software? I have been searching
for ABX for Mac.


/Peter

http://hddaudio.net/wp-content/uploa...3/abxerapp.zip

[Audiop]

Thanks!


/Peter

[Petrus]

I used Foobar ABX comparator on these files and could not tell them apart.

Out of curiosity I also tested a 24/96 classical violin concerto file to a 16/44.1 version (Sox SRC with Shibata dither to 16/44 then back to 24/96 with Sox SRC) and could not tell them apart: results were 11/20 and 12/20 which is guessing accuracy. The original recording was done in DXD.

Next I inverted the phase of one of the files and summed them in Reaper. There are some conversion artefacts at -120 dBFS level scattered between 0 and 15000 Hz (these are well below the thresold of hearing at normal listening levels) and dither hump near 20 kHz, average RMS about -70 dBFS. The difference signal can not be heard untill the monitoring level is turned up about 10 dB over standard K-20 calibration level.

Good enough for me: I cannot hear the difference between 24/96 and well converted 16/44.1. Case closed.

Still I continue recording and editing at 24/88.2 and do the SRC with Sox or iZotope as the last step before import to CD Architect. This is just for the peace of mind and in case the client wants "hi-res files" for some purpose.

[Key]

Originally Posted by Petrus

Out of curiosity I also tested a 24/96 classical violin concerto file to a 16/44.1 version (Sox SRC with Shibata dither to 16/44 then back to 24/96 with Sox SRC) and could not tell them apart: results were 11/20 and 12/20 which is guessing accuracy. The original recording was done in DXD.

Next I inverted the phase of one of the files and summed them in Reaper. There are some conversion artefacts at -120 dBFS level scattered between 0 and 15000 Hz (these are well below the thresold of hearing at normal listening levels) and dither hump near 20 kHz, average RMS about -70 dBFS. The difference signal can not be heard untill the monitoring level is turned up about 10 dB over standard K-20 calibration level.

With 48kHz and higher you can put the dither in the 20kHz-nyquist range. Now am I wrong but from my testing it looks like if you can put the dither out of the audible range you can get an effective 24-bit SNR with 16-bits.?.?

Anyway I am pretty sure I would not be able to tell the difference in this test as well even if I was supplied with music - I can do it myself I guess since ABX is always blind. But this is where I start to become a perfectionist with what I want to spend my money on or purchase. Right now I can get a 2TB hard drive for what? A little more than a $100? When space is that cheap do I really need to be squeezing every little bit out of the music? Or if I can afford it would I pay a little bit more for a 192k 24-bit or 96k 24-bit copy of an album I love? At the rate these hard drives are getting smaller I wouldn't doubt that you can fit more mp3s than you could actually listen to in your lifetime on one. So give me a big old clunky hi resolution flac file anyday - I will do that perceptual encoding and bit crunching stuff myself because I don't really trust other people to do SRCs and lossy encodings for me. I guess I am a sucker for the concept of "as close to the original as possible" even if it is superfluous and I probably can't hear the difference.

[JoeDeF]

Originally Posted by Luna Sound

http://hddaudio.net/wp-content/uploa...3/abxerapp.zip

Thanks, Luna Sound, for the ABXer link. Has anyone been able to download it? I haven't gotten the link to work. I went ahead and registered, but I still can't download it. Has it perhaps been deleted from their server?

If anyone knows how to download this app from hddaudio, or of another place from which ABXer can be downloaded, I'd appreciate knowing about it.


Thanks,
Joe

[Luna Sound]

Originally Posted by JoeDeF

Thanks, Luna Sound, for the ABXer link. Has anyone been able to download it? I haven't gotten the link to work. I went ahead and registered, but I still can't download it. Has it perhaps been deleted from their server?

If anyone knows how to download this app from hddaudio, or of another place from which ABXer can be downloaded, I'd appreciate knowing about it.


Thanks,
Joe

It was working when I posted. Unfortunately you're right that it doesn't seem to any more... sorry.

[AudioJota]

Hi everyone, been lurking for a while and decided to get a little involved.

I ABXed with foobar and got 7/10, I only tried with the shaker and I have to say that by the last four, I was desperately wanting it to end.
I liked SRCc more than SRC001, I feel it had something like a thicker impact in the lower part of the spectrum and it was kind of denser in the richness of the tiny parts moving. Kind of hard to put into words huh!
IMHO SRCc was a little better

[Audiop]

Thank you for participating AudioJota.


/Peter

[AudioJota]

Originally Posted by Audiop

Thank you for participating AudioJota.


/Peter

Sounds like I lost!

Thanks for welcoming

[rogerbrain]

Originally Posted by Bob Olhsson

FWIW high frequency reproduction is not why one uses high sample rates.

that seems to be where the 'quest' starts tho....

frequency response. filter artifacts, 'native' SR in converters, clocking error and jitter, Summing all seem to play a part

it would seem that the correct choice is not one size fits all but what works best on the converters you are using in your DAW of choice ..I have given up trying to find and quantify the data on my rig and just use 192... sound pretty good.. computer power and storage are not issues for me so..why not?