Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Ivo]

Jon,

all I can say is that the same instrumental samples recorded alternately in 96 and 44k do sound different (if one listens by switching between the two original sessions). Resampling 96 k to 44k retains most of the 96k sound characteristics that the sound differences are still audible.
Yes, we are listening in 44k now - but within that frequency there is a whole universe of all possible sounds ...

[Jon Hodgson]

Ivo wrote on Wed, 21 December 2005 00:34

Jon, all I can say is that the same instrumental samples recorded alternately in 96 and 44k do sound different (if one listens by switching between the two original sessions). Resampling 96 k to 44k retains most of the 96k sound characteristics that the sound differences are still audible. Yes, we are listening in 44k now - but within that frequency there is a whole universe of all possible sounds ...

Ivo,

I'm not denying the possibility that things will sound different if the output of the Lavry converter is 96kHz rather than 44kHz, they certainly will BE different, since the noise floor at 96kHz is slightly higher, what other differences there are between the two within the audio band I don't know.

But when you compare two different signals, one of which was recorded directly at 44kHz, and one at 96kHz then converted down to 44kHz, then you have three main variables.

1) You are comparing two different signals... sorry but you've got a problem right there (I know it's difficult to avoid, in fact the best way would require internal knowledge of the converter itself and some DSP code).

2) You are comparing TWO DIFFERENT 44kHz samplers.
a) The Lavry fed a direct analogue input
b) The SRC fed a digital input generated by the Lavry.

3) By pre-declaring which is which you also bring in personal bias into the equation. Give me a room of people and I could convince at least some of them that two identical signals were quite obviously different.

If we ignore the last factor in this case, since the signals are measurably different, you still have the first two variables.

You asked the question why the Lavry sounds better at 96kHz than 44kHz, well the simple fact is that if these "improvements" (nicer is not always more accurate) translate through SRC then it is not down to there being 96000 rather than 44100 samples.

If you could send me the original samples (PM me for email) then maybe we could investigate further.

[Ivo]

Jon,

thank you for your reply. I am far from being any "digital expert", I am just a practical musician following the golden audio rule: if it sounds good, it is good. For sure, I can post also the original 96k samples (or maybe part of them due to their size). I will do it in the evening
Looking forward to the further inspiring and revealing discussion

Sincerely,

[Jon Hodgson]

Ivo wrote on Wed, 21 December 2005 09:10

Jon, thank you for your reply. I am far from being any "digital expert", I am just a practical musician following the golden audio rule: if it sounds good, it is good.

You're quite right. But job of engineers and scientists (and this is an engineering and science forum) is to work out WHY it sounds good, once we know that we work to improve or control this particular element.

Simplistic thinking is fine when you're dealing with what is in front of you, so if you prefer the sound of your Lavry converters sampling at 96kHz and then SRCd down to 44kHz, then that is a valid artistic choice. But if you or others jump to the conclusion as to what the cause is, and therefore extrapolate from this that sampling at 96kHz is ALWAYS the best first stage (whatever your converter or your source material or whatever other variable), and then use this to predetermine your recording and purchase decisions, then things go awry.

It's like if you take a ride in a Mercedes that does 160mph and feels really comfortable at 70, and jump to the conclusion that to be comfortable at 70 you need a car which can go at 160... so then you go out and buy a sports car not realizing it has rock hard suspension and will actually be less comfortable at 70 than your family hatchback.

Ivo wrote on Wed, 21 December 2005 09:10

For sure, I can post also the original 96k samples (or maybe part of them due to their size). I will do it in the evening Looking forward to the further inspiring and revealing discussion Sincerely,

I may be able to give you some ftp space to upload them to, I have a whole server at my disposal but I'm still working out how to operate it.

[bobkatz]

Jon Hodgson wrote on Wed, 21 December 2005 02:20

I'm not denying the possibility that things will sound different if the output of the Lavry converter is 96kHz rather than 44kHz, they certainly will BE different, since the noise floor at 96kHz is slightly higher, what other differences there are between the two within the audio band I don't know.

All other things being equal, the signal to noise ratio in the audible band goes up 3 dB when you halve the bandwidth during an SRC. In other words, the noise floor in the 0-20 kHz band of the 96 K file is half that of the noise floor of the 0-40 kHz band.

The other variables are far too great to have a definitive answer why one sounds better. I go for the filters, but you already know that  

Bk

[Mark Lemaire]

Ivo-

Thanks for your informative postings and useful samples. Before I tell you my impressions of the converters etc let me thank you for the pleasant music- lovely acoustic sounds.

I am considering upgrading my converters and the Lavry is on my list, as well as considering whether to stop recording at 44.1 and start doing 96k or whatever. I typically consider it useless to record at a high sample rate unless one can hear the results at the final 44.1 rate, so your samples that in final form have been SRC'd to 44.1 suit my purposes well.

Anyway, I downloaded your samples and then, in order to create a blind listening test, relabeled the files with random names so that I would not know the original sample rate-- but I kept a 'key' so I could check my results later. Then I listened to the samples and wrote down my impressions in 2 seperate listening tests.

And my results?

On the drum samples, my results were seemingly random- on both tests I chose either the 44.1 or the 96k samples as 'better' or 'more air' with no correlation as to what the orig sample rate was.

On the Viola tests I disqualified the Vl#1 samples because I had 'which was which' memorized before setting up the blind test. On the blindfold Vl #2 test, I chose the 96k sample as 'more closed" and the 44.1 sample as 'more edge' or 'more top' both times (!)

Monochord: On Mono#1 I chose the 96k sample as having 'more open' and 'better seperation' than the 44.1 sample in both tests. But on the mon #2 samples, I chose the 96k sample as 'more top & seperation' the first test, but then labeled it 'darker/ less air' the second test.


All this means that to my ears the samples are functionally about the same, and there is no need (yet) for me to start using up more HD space to achieve better sound.

Thank you for the opportunity to test my ears this way!

I'd love to hear from others who have set up a blind test for themselves and what they think of the results. I have great trouble trusting my impressions of subtle shifts in audio quality without a blind test.

thanks!

Mark Lemaire

[maxdimario]

Quote:

But Max, These are ALL 44.1kHz samples. So why do they sound different? It's NOT the sample rate, because it is the same.

jon,

listen to the bass test samples for now, they are still around I think.

Ivo, please post two short uncompressed sound-files one at 44 and the other at 96 so that we can avoid any arguments.

I'd love to do it, but you have access to a top lavry converter and you have some wonderfully rich-sounding musical instruments.

finally we can end these discussions.

anyway, I've noticed from previous posts that compressed files seem to suffer more than uncompressed files, which makes it even MORE of an issue in today's lo-fi MP3 world...strangely enough!

[maxdimario]

I was listening on headphones through my computer output (it's a quosmio so it's a slight bit better than average) and the difference was neat and clear-cut.

no mistery to me.

[Jon Hodgson]

maxdimario wrote on Thu, 22 December 2005 00:20

I was listening on headphones through my computer output (it's a quosmio so it's a slight bit better than average) and the difference was neat and clear-cut. no mistery to me.

Max,

I'm guessing that your theory as to the explanation of the difference is that sampling at 96kHz is superior to sampling at 44kHz.

But you're comparing two different 44kHz samplers, not a 44kHz and a 96kHz one.

So if there is an improvement, then what it tells you is that for your purposes (since "nicer" is subjective) you prefer the effect of the filtering and quantization in his SRC to the effects of the final decimation filter in the Lavry converter.

[Ronny]

Mark Lemaire wrote on Wed, 21 December 2005 16:33

Ivo- Thanks for your informative postings and useful samples. Before I tell you my impressions of the converters etc let me thank you for the pleasant music- lovely acoustic sounds. I am considering upgrading my converters and the Lavry is on my list, as well as considering whether to stop recording at 44.1 and start doing 96k or whatever. I typically consider it useless to record at a high sample rate unless one can hear the results at the final 44.1 rate, so your samples that in final form have been SRC'd to 44.1 suit my purposes well. Anyway, I downloaded your samples and then, in order to create a blind listening test, relabeled the files with random names so that I would not know the original sample rate-- but I kept a 'key' so I could check my results later. Then I listened to the samples and wrote down my impressions in 2 seperate listening tests. And my results? On the drum samples, my results were seemingly random- on both tests I chose either the 44.1 or the 96k samples as 'better' or 'more air' with no correlation as to what the orig sample rate was. On the Viola tests I disqualified the Vl#1 samples because I had 'which was which' memorized before setting up the blind test. On the blindfold Vl #2 test, I chose the 96k sample as 'more closed" and the 44.1 sample as 'more edge' or 'more top' both times (!) Monochord: On Mono#1 I chose the 96k sample as having 'more open' and 'better seperation' than the 44.1 sample in both tests. But on the mon #2 samples, I chose the 96k sample as 'more top & seperation' the first test, but then labeled it 'darker/ less air' the second test. All this means that to my ears the samples are functionally about the same, and there is no need (yet) for me to start using up more HD space to achieve better sound. Thank you for the opportunity to test my ears this way! I'd love to hear from others who have set up a blind test for themselves and what they think of the results. I have great trouble trusting my impressions of subtle shifts in audio quality without a blind test. thanks! Mark Lemaire

Mark, let me commend you on your testing methods and your observations. If everyone were using similiar techniques, we'd have a lot less differences of opinions and much less voodoo science theories. While it's better to do blind tests with another participant, you've devised a way to eliminate expectations leading to imagination on a solo test. You are on a good track not trusting sighted impressions.


The more you test in the blind, the more you will realize how easy the ear can be influenced by the other senses, in particular sight. The mind by expectations, mood, fresh ears, fatigued ears, how ceruminous an individual's ears may be, and what previous level the ears are accustomed to, at any given minute of any day.


There are so many variables when trying to determine sample rate differences, that it's pert near impossible when high end gear is used, to pinpoint it down to just the sample rate. Even the best, blind tests, are never perfect and they always seem to pose more questions than they answer, but IMHO, they are the best we've got. It pays to confirm your non-sighted impressions with blind evaluations. If there are significant differences, they will be detected in the blind with more validity than an evaluation that starts with expectations, IMHO.  

[squeegybug]

I far preferred the additonal depth and resonance of the 96 KHz samples.

If, from some opinions I'm reading here, that is simply the result of resampling errors or anomalies, then sign me up for some of that flawed SRC program.

How about this -- maybe the 96k conversion sounds better because... it's better.

For additional amusement or speculation, here are two clips I recorded.  Two separate performances: One at 96 KHz/24 bit, the other at 48 KHz/24 bit.  Both resampled and dithered to 44.1/16:

File_1
File_2

And here is the raw 96/24 version.

Steve

[Jon Hodgson]

squeegybug wrote on Thu, 22 December 2005 07:39

How about this -- maybe the 96k conversion sounds better because... it's better.

That's not an answer that works on a science and engineering forum. The rules of this forum is that we talk facts, not subjective assessments - naturally subjective assessments may lead us to investigate and discover facts, but we don't just stop at the point of "it sounds nicer".

If the improvement was specifically due to the higher sampling rate (as opposed to some other variable which changed, either because in the case of the Lavry converter it is linked to the sample rate, or by chance), which would mean that frequencies above the cutoff point of the final anti-aliasing filter were involved (basically over 20kHz), then this would not translate
through the sample rate conversion, since no matter what you do there is no way that a 44kHz sample stream can contain any information over 22kHz.

I have a couple of theories, but I'll need the original samples to investigate them

[squeegybug]

Jon Hodgson wrote on Thu, 22 December 2005 02:30

squeegybug wrote on Thu, 22 December 2005 07:39 How about this -- maybe the 96k conversion sounds better because... it's better. we don't just stop at the point of "it sounds nicer".

Yes, I appreciate that.  My point was directed to the discussions that the SRC was somehow changing the original source, even to the point of "improving" it (to my ears, subjective I know).

Whereas I believe, but cannot prove except by experiencing, that the native higher rate files are somehow simply a more accurate representation of the live performance.  And my converters and SRC are not the same as Ivo's.

So that is where I would expect the discussion to focus -- as the OP asked, why do the higher rates sound better?  But I would also think this is a question that has surely been asked since the beginning of analog to digital conversion?

Steve

[Jon Hodgson]

squeegybug wrote on Thu, 22 December 2005 09:38

Whereas I believe, but cannot prove except by experiencing, that the native higher rate files are somehow simply a more accurate representation of the live performance.  And my converters and SRC are not the same as Ivo's.

This is possible, if the additional bandwidth were more important than the signal to noise ratio...

However what is NOT possible is for this additional accuracy to survive a SRC unscathed. a given stream of samples (sample rate and bit depth) is always capable of containing exactly the same amount of audio information versus noise, WHATEVER THE ROUTE USED TO GENERATE THAT STREAM.

Now it is perfectly possible for a converter design to have certain inadequacies which mean that it is better to take its output at 96kHz and then downconvert it using an additional unit or software (that's actually what is going on inside the converter anyway), but this would be a result of the physical implementation of that circuit or the internal processing, and not some magical circumvention of Nyquist.

[squeegybug]

Jon Hodgson wrote

However what is NOT possible is for this additional accuracy to survive a SRC unscathed. Now it is perfectly possible for a converter design to have certain inadequacies which mean that it is better to take its output at 96kHz and then downconvert it using an additional unit or software (that's actually what is going on inside the converter anyway), but this would be a result of the physical implementation of that circuit or the internal processing, and not some magical circumvention of Nyquist.

And of course that slight, but obvious, degradation from the SRC is apparent between the original 96 and the 96-resampled-to-44.1 files that I posted, and Ivo reports similar observations.

I find it interesting that our different converters and SRC still both produce superior (subjective) results at the higher rate.  I guess the mathematical hunt here is for any common ground that will explain this... or perhaps this would actually prove to be the case with all converters, except those with obvious deficiencies at high or low rates.  Sounds like a difficult project.

Steve