'Normalization' of PCM audio - subjectively benign? Options

[greynol]

I think RockFan bowed out of this discussion.

After checking Hydrogenaudio Forums > Misc. > Recycle Bin, I think I know why.

This post has been edited by greynol: Sep 6 2006, 17:29

[pepoluan]

Quite rare seeing TOS#2 got invoked

[cabbagerat]

Cool, thanks uart! What brand of scope is that? Digital scopes are really great tools, but they just don't have the same feel as the old school green phosphor CRT type

[krabapple]

2bdecided wrote:
Searching Google for dither (here in the UK at least - Google results are regionalised, even if you select all of the web)...


The 3rd hit is by Nika Aldrich. Nika turned up on one forum years ago (I can't remember if it was r3mix, mp3.com, or somewhere else) proudly announcing his new article on dither. Let me say first that, in his area of expertise, Nika Aldrich is widely respected. However, dither apparently wasn't his area of expertise at the time, and his article was roundly criticised for being simply incorrect. From the advice on that forum (which would have included advice from people who are on HA now) I believe he corrected his article.

He did -- it's noted with a date at the end of the article. I believe his book (Digital AUdio Explained for the Recording Engineer) incorporates the correct information too (and has a thank-yous section)

This article is now probably very useful, because it is written from the point of view of someone learning about dither.

It was to me.

However, I can't bring myself to like it simply because I remember how wrong it was in its first draft, and how many of the r3mix/mp3.com/HA crew are the true authors, and don't get any credit.

Having corresponded with and talked to Nika, it's hard for me to imagine he would purposely deny them credit if due. Have any contacted him to complain?

The 8th hit - finally - is by Bob Katz. Now, this guy is a genius.

http://www.digido.com/portal/pmodule_id=11...der_page_id=27/

Read that one. Not the first 7. It's written by someone who knows exactly what they're talking about, it includes some pictures, and he's a very nice guy too (he helped me with ReplayGain).

Yes, I've had a couple of nice email interactions with him as well. I like his book very much too (Mastering Audio). I do wish he would publish some of his blind/ABX test results though, given some of the things he claims are audible.

Every attempt to state what should be obvious - that our ears and how enjoyable and 'realistic' music playback is (or isn't) should ultimately arbitrate on music reproduction is met with the catch-all "prove it. Show your ABX results", and TOS invoked. Unfortuntately this leaves little room for any meaningful debate.

You can enjoy your music as and when you please.

The purpose of ABX is to demonstrate that an audible difference exists.

Not to mention that blind tests are the best known way to make sure one's *ears* are 'arbitrating', rather than one's eyes or one's prejudices. If someone's not willing to do a blind comparison, maybe they don't really trust their ears.

This post has been edited by krabapple: Sep 6 2006, 20:46

[pepoluan]

Not to mention that blind tests are the best known way to make sure one's *ears* are 'arbitrating', rather than one's eyes or one's prejudices. If someone's not willing to do a blind comparison, maybe they don't really trust their ears.

If they don't trust their ears, perhaps they need to use some DSPs

[db1989]

Let's just hope he's gone for good. uart has proved what everyone else knew. Thanks, uart!

[2Bdecided]

(images snipped)
Top Trace : 14khz tone as 44.1k 16bit PCM file. This is the view of the actual file when loaded into "Audacity".

Middle Trace : Actual analog out when above PCM file is played (foobar) without any resampling or DSP. Soundcard is onboard "soundmax" AC97 of cheap motherboard. (Photo of osciloscope image)

Bottom Tave : As above but zoomed in.

This bit (with images) should be in the FAQ!

The fact that it even works with an on-board AC97 sound card says it all!

(Mind you, it's probably resampled to 48kHz internally, though that's not relevant to the point you're proving).

Cheers,
David.

[uart]

Cool, thanks uart! What brand of scope is that? Digital scopes are really great tools, but they just don't have the same feel as the old school green phosphor CRT type

Hi cabbagerat. It was my brothers osciloscope that I'd borrowed last weekend for another project. I cant recall the model number right now but I think it's only a cheapie. Anyway I was reading this thread and the silly claims made on that page that Rockfan linked to so I thought I should snap a view photos of actual waveforms to help debunk it. I didn't get around to uploading them for a few days but I'm glad I did. I know it just confirms what most of us already knew anyway.

BTW. The small artifacts that you can see on those osciloscope traces are properties of the oscilscope and not the waveform. If rockfan is still reading I can assure him that if I connect an analog signal (sine) generator to the same scope that the waveforms look exactly the same.

Well as far as the "16bit 44.1kHz PCM is not good enough for really high quality audio" debate goes, I'm not totally closed minded to the possiblity that higher bit depths and sample rates might offer some small improvement to some people, but I really hate to see CD digital audio unfairly criticized like in that above link. Personally I am somewhat skeptical of whether the higher bitrate stereo (like 96kHz 24 bit etc) will ever be consistantly ABX'able over a well mastered CD. Sure it's a great idea to master the music at higher bitdepth and samlpe rate. Higher sample rates make the analog anti-aliasing filters much easier and higher inital bitdepths mean that edits can be lossless at the final bitdepth level. I think they choose the standards pretty well with the good old CD. They certainly the didn't go overboad on the bitrate side, with only 650MB per disc they couldn't afford to, but 44.1,16 still gives 20khz bandwidth and about 100dB dynamic range and damit that's good enough for me.

[greynol]

Higher sample rates make the analog anti-aliasing filters much easier

Aliasing occurs on the A/D side of things, not the D/A side of things.

I think the misuse of this word was corrected a while back.

EDIT: A while back in a different thread, sorry.
http://www.hydrogenaudio.org/forums/index....&pid=427756

Of course a higher sample rate will require a more simple prefilter but I'm pretty sure you're talking about playback, not sampling.

This post has been edited by greynol: Sep 7 2006, 18:03

[uart]

Aliasing occurs on the A/D side of things, not the D/A side of things.
I think the misuse of this word was corrected a while back.

Yes that's what I meant. When I said mastering I really meant "recording and mastering", the A/D side was most definitely what I had in mind when I mentioned anti-aliasing filters.

The quote in context was

Sure it's a great idea to master the music at higher bitdepth and samlpe rate. Higher sample rates make the analog anti-aliasing filters much easier and higher inital bitdepths mean that edits can be lossless at the final bitdepth level.

Maybe I didn't really make it clear what I was trying to say. I was trying to say that I can see definite advantages to recording and mastering at higher bit-rates but at this point in time I'm still somewhat skeptical about whether higher bit-rates (96k 24bit etc) will offer any real advantage for stereo playback. Not for most listeners in any case.

This post has been edited by uart: Sep 7 2006, 18:16

[greynol]

Sure it's a great idea to master the music at higher bitdepth and samlpe rate. Higher sample rates make the analog anti-aliasing filters much easier and higher inital bitdepths mean that edits can be lossless at the final bitdepth level.

[krabapple]

Check out this spin on the 'square wave' 'problem'.
(It's Chris(tine) Tham again ...who used dubious methods to show that LPs have more dynamic range than CDs, in another Audioholics article)

http://www.audioholics.com/techtips/specsf...igitalAudio.php

scroll to the bottom part about 'non-sine waves'

This post has been edited by krabapple: Sep 7 2006, 20:54

[greynol]

I read your post regarding the conclusion of that link prior to editing.

This is perhaps true, but the counter argument is that if a digital player is not playing back a 0dB 19997kHz sawtooth (or a 0dB 10kHz square wave) with even amplitude accuracy (let alone harmonic accuracy), then that represents a kind of “distortion” that is audible.

It is hillarious how the author can talk about "amplitude accuracy" without bothering to say anything about quantization error. What is being talked about here is solely "harmonic accuracy."

This post has been edited by greynol: Sep 7 2006, 21:12

[SebastianG]

The article started out pretty good. But just to make sure that nobody gets the wrong idea: She made a big mistake.

• She says something about Nyquist. Good for her. (Though she should also mention that an anti-alias filter before sampling is needed. She only concentrates on the reconstruction. Maybe she doesn't know about aliasing.)
• She seems to be aware of the fact that a square wave contains many high frequency harmonics that are above the Nyquist frequncy. Funny thing is: She says sine waves can be reconstructed properly but non-sine waves not. Contradiction alarm!
• She sampled a non-bandlimited 10 kHz square wave thus violating Nyquist to prove her point.

This post has been edited by SebastianG: Sep 8 2006, 13:28

[greynol]

Her square wave was produced within Adobe Audition after choosing 44.1kHz/16-bit for the settings. I get exactly the same shape as what is shown in the article.

[SebastianG]

Are you implying something?

EDIT:
Because a 10 kHz square wave is composed out of the following harmonics: 10 kHz, 30 kHz, 50 kHz, 70 kHz, 90 kHz, .... the only thing you should get at 44.1 kHz sampling rate is a 10 kHz sine wave. Assuming you're right (I don't have Adobe Audition to test it) you may want to check the spectrum view. I'm sure you'll see a lot of aliasing. ==> Don't use Adobe Audition for generating other-than-sine waves as it samples them without removing harmonics above the Nyquist frequency first.

EDIT2:
You would expect Adobe Audition to create proper waveforms. She probably trusted Adobe Audition. Thou shall not jump to conclusions without knowing exactly what the software does you are working with. This applies to all the RockFan-type people. In his case he was fooled by a software's wave form visualization.

EDIT3: I just tested Audacity. You'll get horrible aliasing as well when you generate sqaure waves. Google for alias-free waveform generation and/or bandlimited step if you're intestested in digital sound synthesis.

This post has been edited by SebastianG: Sep 8 2006, 13:31

[greynol]

That she knew the square wave that she produced was bandwidth limited.

The label of the plot was "Figure 9: 10kHz 0dB square wave sampled at 44.1kHz 16 bits"

Either this or I fail to grasp your point. I mean, what would a non-bandlimited 10kHz 0dB square wave look like except for what was in the plot?

EDIT: Ah, yes (ligthbulb goes on), a prefiltered 10kHz square wave would be a 10kHz sine wave. I failed to grasp your point!

This post has been edited by greynol: Sep 8 2006, 09:09

[2Bdecided]

Cool Edit Pro (so presumably audition) is useless for alias-free square waves

10kHz generated at 44.1kHz...

Waveform:

Spectrum:

If you listen to it, it sounds like the fundamental is at 100Hz, with lots of screeching on top!

TBH, you would have to be a bit daft to listen to the result and still think that the software was working properly.

When I first spotted the problem, I "solved" it by generating square waves (actually swept square waves) using Cool Edit Pro at 100x the target sample rate (i.e. 4410kHz sampled) and resampled them back down to 44.1kHz. It gets rid of most of the false harmonics, but there's still a few ~30dB down.

Thinking about it, there are neater solutions, even in Cool Edit. e.g. generate the required (in band) harmonics manually by summing sine waves (settings in the tone generator let you do this).

Cheers,
David.

[uart]

Thinking about it, there are neater solutions, even in Cool Edit. e.g. generate the required (in band) harmonics manually by summing sine waves (settings in the tone generator let you do this).

Yep, that's exactly what Sebastian was pointing out. Both of those "problem samples", the 10kHz square wave and the 19.997kHz triangular wave, would each precisely be a perfect sine wave if they were band-limited to 20KHz before sampling as they should have been. The only relevant question that the author could legitimately have posed would have been whether the human ear could distinguish between those waveforms and the corresponding filtered sine wave when played though high end stereo equipment. Certainly the author made no attempt at all to demonstrate that she could hear the difference between a 19.997kHz triangular wave and a 19.997kHz sine-wave.
.

About Christine Tham
Christine Tham has always been a keen "hi fi" enthusiast, which is an affliction she inherited from her father. She was interested enough in the subject to enroll in an Electrical Engineering degree at Sydney University , but decided that was not for her and graduated instead with a University Medal in Computer Science (Honours) and subsequently a Master of Applied Finance from Macquarie University.

Hey, too bad that she pulled out of that EE course before they covered the stuff pre-sample filtering.

[pepoluan]

Hey, too bad that she pulled out of that EE course before they covered the stuff pre-sample filtering.

And not even all EE programs has Digital Signal Processing... case in point: There are 7 majors in the EE program of my university, and only 2 of them requires the DSP courses; optional for the other majors...

Edit: Added "my university" there.

This post has been edited by pepoluan: Sep 8 2006, 19:21

[Axon]

When I went to college - for an EE degree - the DSP class was optional. And I didn't take it. I really wish I did, but as it turns out, one of the books I wound up having had a very good treatment of the topic, so I've studied up.

Once you have a good grasp of Fourier transforms you can figure the rest out. Unfortunately, only EEs (well, and probably MEs and physics majors) ever have a need to delve into that stuff.

This post has been edited by Axon: Sep 8 2006, 18:24

[pepoluan]

Yes, all majors in EE requires a knowledge of Fourier transforms.

However I finally recall that Discrete Signal Processing (in my university) is mandatory only for the Telecommunications major. It is optional for all the other majors... and due to its complexity, naturally no one wants to take that satanic course

[krabapple]

I posted a complaint thread about Tham's stuff on Audioholics. Any takers here for the reply from the editor?

http://forums.audioholics.com/forums/showt...6095#post206095

Perhaps you can direct the more knowledgable people on those forums to construct a new test sequence that is more real world? We're open to additional input.
__________________
Clint DeBoer
Editor in Chief
Audioholics

This post has been edited by krabapple: Sep 8 2006, 20:08

[jlt]

We're open to additional input

why he don't came here as here was started?
we're open too(and a long time)

back to the title: 'Normalization' of PCM audio - subjectively benign?

i have read about lossy and lossless,lossy lose few details and lossless don't lose anything.
seems ok!
imagine if one (sample)music that we don't know what was done,if normalized,or equalized or if was changed from 16 to 32bit and back to 16bit again,or was 44.1k to 48k and back to 44.1k again,etc.

@ moderators
can i host the (sample)music for analysis?

if i can host and you all download,as nobody knows if any effect was used,i ask:

1- was normalized(how much?) or was not normalized?
2- if was normalized,is benign or not?
3- is 44.1k now but was 48k?
4- if was 48k and now is 44.1,what is lost?
5- if was 48k and now is 44.1k,how was dithered?
6- was really dithered? how?
5- was equalized or not?
6- if was equalized,where and how much was done?
..of course,we can encrease the numbers of questions.

if anyone can tell what was done with the source that i want to host,i can trust that everybody can tell if normalization is benign or not and if have real audibles differences between lossy and lossless.
if nobody can't hear,what is the difference if normalization is benign or not?

the central point: is possible to have one true answer?

thanks

[Axon]

I'll take on the dynamic range article if somebody else comments on the sampling article.

EDIT: I replied to that thread, comments on my analysis welcome.

This post has been edited by Axon: Sep 9 2006, 00:56