IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
Vorbis-encoded white noise contains info well above 20kHz?
fedetxf
post Oct 23 2009, 22:07
Post #1





Group: Members
Posts: 30
Joined: 17-August 04
From: Argentina
Member No.: 16354



Here's the wikipedia's sample white noise ogg vorbis file.

http://en.wikipedia.org/wiki/File:Whitenoisesound.ogg

I opened Audacity and analized its spectrum. To my surprise, it is complete up to 22 kHz (being a 44.1 kHz sampled file). Usually vorbis music files contain frecuencies up to somewhere 16 to 18 kHz. Removing high frecuencies is parto of lossy compression and this vorbis file is not q10 or anything. But I think even q10 files don't have anything above 20 kHz.

For example this vorbis file

http://en.wikipedia.org/wiki/File:Johann_S...02Epre_cmaj.ogg

doesn't have anything above 20 kHz and very little above 18 kHz. And this one:

http://en.wikipedia.org/wiki/File:Bach_-_B..._1._Allegro.ogg

doesn't have anything above 10 kHz.

Maybe one of you with deep vorbis knowlegde know the explanation or can tell me why I'm expecting something that I should not expect. Maybe I'm using Audacity in the wrong way or misinterpreting it someway.
Go to the top of the page
+Quote Post
pdq
post Oct 24 2009, 00:11
Post #2





Group: Members
Posts: 3375
Joined: 1-September 05
From: SE Pennsylvania
Member No.: 24233



Unless you encoded the files yourself and know EXACTLY what command line was used, I wouldn't pay too much attention to this.
Go to the top of the page
+Quote Post
DVDdoug
post Oct 24 2009, 00:46
Post #3





Group: Members
Posts: 2539
Joined: 24-August 07
From: Silicon Valley
Member No.: 46454



I don't have any "deep knowledge", an I dont use OGG. But, I know these lossy compression schemes are based on psychoacoustics, not visual spectrum analysis.

If both the music and the white noise sound like the uncompressed original, the compression is doing its job!

White noise has relatively more high frequency content than most music. If there is lots of "important" or dominant high-frequency information, I'm not surprised if the encoder tries to preserve the dominant content. The algorithm trys to preserve the most important information and in some cases, with "simple" sounds, the algorithm may be able to compress the sound without actually throwing-away any information. (I assume that when you compress a constant 1kHz tone with lossy compression, nothing is lost!)

In theory, it's possible to "perfectly" compress a pure 20kHz tone (or higher) at very low bitrates.* But, the compression algorithms are optimized for music, not pure tones, and I don't have the "deep knowledge" to know how well the various algorithms handle pure tones.

Just listening to the Bach pieces, the first sounds relatively "high pitched", but I don't hear a lot of high frequency energy. The second one is more balanced, but with less high frequency content. To me, it doesn't sound like "bad recordings" or bad encoding,... It just sounds like the nature of the classical (baroque?) instruments. If there were cymbals or brass, there would be more high-frequency content. I don't know, but it's possible that you wouldn't see any high frequencies in the spectrum of an uncompressed version of the same recording.


* If I say "20kHz, 0dB, 1 hour", that's all the information you need to reproduce that tone, and I've "compressed" that information into 18 bytes!
Go to the top of the page
+Quote Post
Fedot L
post Jan 11 2010, 00:35
Post #4





Group: Members
Posts: 109
Joined: 2-January 10
Member No.: 76586



QUOTE (fedetxf @ Oct 23 2009, 22:07) *
To my surprise, it is complete up to 22 kHz (being a 44.1 kHz sampled file). Usually vorbis music files contain frequencies up to somewhere 16 to 18 kHz… I think even q10 files don't have anything above 20 kHz.

It firmly depends on the bitrate used. In my experiments, encoding of white noise or real music material rich in high frequencies into Ogg Vorbis at VBR 128 kbit/s gave stably 19 kHz linear bandwidth, at VBR 160 kbit/s 20,3 kHz linear bandwidth, and at VBR 192 kbit/s and higher VBRs, gave stably 22 kHz linear bandwidth displayed on a spectrum analyzer.

You can view all this, if you wish, on a spectrum analyzer, on four files of electronic music (the “original” file uncompressed specially treated to have up to 22 kHz linear spectrum): WAV uncompressed, Ogg Vorbis VBR 128 kbit/s, VBR 160 kbit/s and VBR 192 kbit/s:
http://www.fayloobmennik.net/download.php?...Test_HF_F_1.wav
http://www.fayloobmennik.net/download.php?...F_F_VBR_128.ogg
http://www.fayloobmennik.net/download.php?...F_F_VBR_160.ogg
http://www.fayloobmennik.net/download.php?...F_F_VBR_192.ogg

To download, type four digits you will see into the little box, and press the button at the bottom.

Be careful with volume setting before listening to (if you want to do): VERY HIGH HF LEVEL!!
Better to set the volume to “0”, then very carefully increase it. And with “Treble” tone controls lowered.
Go to the top of the page
+Quote Post
Garf
post Jan 11 2010, 00:42
Post #5


Server Admin


Group: Admin
Posts: 4883
Joined: 24-September 01
Member No.: 13



QUOTE (fedetxf @ Oct 23 2009, 22:07) *
Here's the wikipedia's sample white noise ogg vorbis file.

http://en.wikipedia.org/wiki/File:Whitenoisesound.ogg

I opened Audacity and analized its spectrum. To my surprise, it is complete up to 22 kHz (being a 44.1 kHz sampled file).


1) Most "music" Vorbis files you see don't have much "power" in the high spectrum, or the sounds there are of much lower power than those in the other bands. Due to the combination of ATH, masking and the encoder running out of bits, they'll tend to cut out somewhere around 16-18kHz. White noise has by definition an equal power in all bands and is hence quite atypical. The psymodel will not have any "justification" to remove the HF noise.

2) Quite likely the file was encoded with nonstandard parameters, because I think Vorbis does have a hard lowpass in most modes (it has been a long while since I looked).
Go to the top of the page
+Quote Post
Fedot L
post Jan 11 2010, 13:03
Post #6





Group: Members
Posts: 109
Joined: 2-January 10
Member No.: 76586



QUOTE (Garf @ Jan 10 2010, 23:42) *
1) Most "music" Vorbis files you see don't have much "power" in the high spectrum, or the sounds there are of much lower power than those in the other bands. Due to the combination of ATH, masking and the encoder running out of bits, they'll tend to cut out somewhere around 16-18kHz. White noise has by definition an equal power in all bands and is hence quite atypical. The psymodel will not have any "justification" to remove the HF noise.

2) Quite likely the file was encoded with nonstandard parameters, because I think Vorbis does have a hard lowpass in most modes (it has been a long while since I looked).

To see if it’s right, I tested Ogg Vorbis encoding a music sound material having an up to 22 kHz spectrum where the signal level at 10-22 kHz was 36 to 48 dB lower (!) than at 0 to 3 kHz. At 350 kbit/s and 500 kbit/s, Ogg Vorbis nevertheless restored surely the bandwidth up to 22 kHz.

You can see this viewing the following three electronic music test files (the “original” file uncompressed having up to 22 kHz spectrum) on a spectrum analyzer: WAV uncompressed, Ogg Vorbis VBR 350 kbit/s and VBR 500 kbit/s:
http://www.fayloobmennik.net/download.php?...1-Test_HF_N.wav
http://www.fayloobmennik.net/download.php?...F_N_VBR_350.ogg
http://www.fayloobmennik.net/download.php?...F_N_VBR_500.ogg

To download, type four digits you will see into the little box, and press the button at the bottom.
Go to the top of the page
+Quote Post
Garf
post Jan 11 2010, 15:02
Post #7


Server Admin


Group: Admin
Posts: 4883
Joined: 24-September 01
Member No.: 13



QUOTE (Fedot L @ Jan 11 2010, 13:03) *
To see if it's right, I tested Ogg Vorbis encoding a music sound material having an up to 22 kHz spectrum where the signal level at 10-22 kHz was 36 to 48 dB lower (!) than at 0 to 3 kHz. At 350 kbit/s and 500 kbit/s, Ogg Vorbis nevertheless restored surely the bandwidth up to 22 kHz.


So, I think your conclusion is that at high bitrates Vorbis doesn't lowpass and is able to keep essentially the entire spectrum (because there are plenty of bits and/or the psymodel parameters are set very paranoid). Correct?
Go to the top of the page
+Quote Post
Fedot L
post Jan 11 2010, 16:34
Post #8





Group: Members
Posts: 109
Joined: 2-January 10
Member No.: 76586



QUOTE (Garf @ Jan 11 2010, 14:02) *
QUOTE (Fedot L @ Jan 11 2010, 13:03) *
To see if it's right, I tested Ogg Vorbis encoding a music sound material having an up to 22 kHz spectrum where the signal level at 10-22 kHz was 36 to 48 dB lower (!) than at 0 to 3 kHz. At 350 kbit/s and 500 kbit/s, Ogg Vorbis nevertheless restored surely the bandwidth up to 22 kHz.


So, I think your conclusion is that at high bitrates Vorbis doesn't lowpass and is able to keep essentially the entire spectrum (because there are plenty of bits and/or the psymodel parameters are set very paranoid). Correct?

But following (viewing), for example, the test files in links, it’s not MY conclusion, but the conclusion made by Ogg Vorbis itself working.

I should never say “paranoid” or “not paranoid” principles of setting psychoacoustical model’s parameters of Ogg Vorbis, by a simple reason: ANY compressed formats encoders sacrifice extreme-high frequencies of sound material as the bitrate lowers, quite naturally. But at 350 kbit/s VBR and 500 kbit/s VBR Ogg Vorbis restores surely the bandwidth up to 22 kHz on any sound material with practically any level of extra-high frequencies. As you can see it.
Go to the top of the page
+Quote Post
[JAZ]
post Jan 11 2010, 20:29
Post #9





Group: Members
Posts: 1765
Joined: 24-June 02
From: Catalunya(Spain)
Member No.: 2383



Ouch... OUch O-UCH!

First, questioning why classical music doesn't have "full" (if we think 22Khz is full) range bandwidth.

Then, showing examples clearly stating that:

QUOTE
Be careful with volume setting before listening to (if you want to do): VERY HIGH HF LEVEL!!
Better to set the volume to “0”, then very carefully increase it. And with “Treble” tone controls lowered.


Are you sure that any real conclusion can be said of Vorbis encoding current Pop/Rock/Top-40/whatever else music, given your "examples"?

Go to the top of the page
+Quote Post
Fedot L
post Jan 12 2010, 00:22
Post #10





Group: Members
Posts: 109
Joined: 2-January 10
Member No.: 76586



QUOTE ([JAZ] @ Jan 11 2010, 20:29) *

…questioning why classical music doesn't have "full" (if we think 22Khz is full) range bandwidth.

As for me, I didn’t say classical music didn’t have "full" 22Khz. “Naturally” performed, it may produce even higher frequencies. But it’s difficult to find recordings of classical music containing valid levels of frequencies above 20 kHz for a valid analysis.
QUOTE ([JAZ] @ Jan 11 2010, 20:29) *

…if we think 22Khz is full range bandwidth.

What do you mean by this? For FM radio “full range bandwidth” is 15 kHz; for stereo vinyl records it’s 20 kHz; for DISCRETE quadra vinyl records it’s 15 kHz; for CD format 16/44,1 it’s 22 kHz etc.
QUOTE ([JAZ] @ Jan 11 2010, 20:29) *

Then, showing examples clearly stating that:
QUOTE
Be careful with volume setting before listening to (if you want to do): VERY HIGH HF LEVEL!!
Better to set the volume to “0”, then very carefully increase it. And with “Treble” tone controls lowered.

Are you sure that any real conclusion can be said of Vorbis encoding current Pop/Rock/Top-40/whatever else music, given your "examples"?

1. I couldn’t see any connection between the citation and your question. And the word “examples” put “in inverted commas”. What do you mean by this?
2. But answering your question “per se”, I’d say “Yes”, the conclusions are quite real. As the tests of very different sound material demonstrate.
Go to the top of the page
+Quote Post
[JAZ]
post Jan 12 2010, 20:11
Post #11





Group: Members
Posts: 1765
Joined: 24-June 02
From: Catalunya(Spain)
Member No.: 2383



Mmm.. I think I didn't fully understand you (I thought it was more in line with the first post).

Anyway, the answers to the questions:

"if we think 22khz is full" -> There is a tendency to expect the whole frequency bandwidth of the sampled signal to be full, be it 32Khz 48Khz or 96Khz. So I was saying I was talking of 44khz sampled signals. (Anyway, it didn't pretend to put too much emphasis on it).


About the citation, my point was that neither the whitenoise, nor the classical recording nor your provided samples were representative of common Vorbis behaviour at the higher part of the frequency spectrum.
Now I see you just wanted to point that Vorbis can record those. I thought you meant to say that it was natural for Vorbis to encode the whole bandwidth.

This post has been edited by [JAZ]: Jan 12 2010, 20:12
Go to the top of the page
+Quote Post
Fedot L
post Jan 12 2010, 23:50
Post #12





Group: Members
Posts: 109
Joined: 2-January 10
Member No.: 76586



QUOTE ([JAZ] @ Jan 12 2010, 20:11) *
About the citation, my point was that neither the whitenoise, nor the classical recording nor your provided samples were representative of common Vorbis behaviour at the higher part of the frequency spectrum.

Mine is opposite, and I consider the samples I cited as fully “representative of common Vorbis behaviour at the higher part of the frequency spectrum”, because at high bit rates 350 kbit/s VBR and 500 kbit/s VBR, Ogg Vorbis demonstrably restores high frequencies up to 22 kHz present in “source” real music material of any level from 0 dB down to -48 dB.
QUOTE ([JAZ] @ Jan 12 2010, 20:11) *
Now I see you just wanted to point that Vorbis can record those.

And it can do.
QUOTE ([JAZ] @ Jan 12 2010, 20:11) *
I thought you meant to say that it was natural for Vorbis to encode the whole bandwidth.

And it is. Maybe we still do not understand each other…

At 500 kbit/s VBR Ogg Vorbis demonstrably linearly restores high frequencies up to 96 kHz (compressed from a 96 kHz -48 dB uncompressed file) even at -48 dB level! But what could be the utility of such capacities? None. Even in case of recording at 192 kHz sampling frequency, applying efforts for to record a bandwidth wider than 22 kHz is of no utility. High and extremely high sampling frequencies serve another purpose: pulses envelope smoothing.

But as for 22 kHz, at high bit rates 350 kbit/s VBR and 500 kbit/s VBR, viewing samples cited demonstrates it at any “source” high frequencies levels.
Go to the top of the page
+Quote Post
[JAZ]
post Jan 13 2010, 20:23
Post #13





Group: Members
Posts: 1765
Joined: 24-June 02
From: Catalunya(Spain)
Member No.: 2383



Then, I have to go strong on my point again. If you need to amplify the higher end of the frequency spectrum to demonstrate that it can, you demonstrate to me that it cannot by default.

Either correct that (by using non-modified music), or there's nothing to talk about.
Go to the top of the page
+Quote Post
Fedot L
post Jan 13 2010, 22:20
Post #14





Group: Members
Posts: 109
Joined: 2-January 10
Member No.: 76586



QUOTE ([JAZ] @ Jan 13 2010, 20:23) *

Then, I have to go STRONG on my point again. If you need to amplify the higher end of the frequency spectrum to demonstrate that it can, you demonstrate to me that it cannot by default.

Either correct that (by using non-modified music), or there's nothing to talk about.

Such a “strength” indicates only that you paid no attention to the whole post #6 developing exactly the results of encoding into Ogg Vorbis “by using non-modified music” having a spectrum up to 22 kHz, without any “amplification of the higher end of the frequency spectrum”:
QUOTE (Fedot L @ Jan 11 2010, 13:03) *
To see if it’s right, I tested Ogg Vorbis encoding a music sound material having an up to 22 kHz spectrum where the signal level at 10-22 kHz was 36 to 48 dB lower (!) than at 0 to 3 kHz. At 350 kbit/s and 500 kbit/s, Ogg Vorbis nevertheless restored surely the bandwidth up to 22 kHz.

You can see this viewing the following… (etc. in the post).
Go to the top of the page
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 30th July 2014 - 22:52