Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Can not tell difference between these samples (Read 26233 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Can not tell difference between these samples

Hi,

i've been doing a some listening tests on Audioexpert comparing different audio types/compression rates and having done some 30 blind tests i've done just a couple of mistakes distinguishing lossy and lossless formats (many of them were quite obvious though due to purposeful (?) noise amplification), where in all the lossy formats anything below 320kbps was fairly easy to distinguish. After that i've decided to do some tests on my own collection.

After that i've decided to do some tests on my personal collection, so i've compared flac to various levels of ogg. And here's the sad news (for me): i could not tell the difference. In ANY case (well, maybe with the exception of levels -1, 0). Actually, i notice an artefact in the ogg file (strangely enough, the higher ogg level - the more amplified the artefact), but apart from it i couldn't really tell the difference between sections of the songs. That leaves me with 3 major options: the flac files were already lossy, that is they were coded using lossy data; i don't have the right tools for comparison; i am "deaf".

The first option falls, since some of data i tested i encoded from cd myself using Exact Audio Copy with referenced parameters.
The second is also unlikely, since i tested both on my laptop and Cowon flac player, using monitoring headphones Sennheiser eh350.
This leaves me with the third option, but i wish to believe otherwise because of audioexpert tests mentioned above as well as really emphasising all-things-music/tools for some years now.

The link to a couple of samples i used. I renamed both ogg files to flac so as to imitate a blind test at least partially (of course you could tell the difference by file sizes alone). First sample is chorus from Ode an die Freude and the other one is plain voice by Roy Harper. Please tell me if you find it easy to distinguish between the samples.

Thank you for your time reading and listening.

p.s. samples were cut using linux shnsplit with manually made .cue file, while encoding was done using oggenc.

Can not tell difference between these samples

Reply #1
So if I understand correctly, you're saying that you can't hear the difference between lossy and lossless files? My take is that it means that ogg is doing its job. Good lossy compressors are supposed to be transparent.

Can not tell difference between these samples

Reply #2
So if I understand correctly, you're saying that you can't hear the difference between lossy and lossless files? My take is that it means that ogg is doing its job. Good lossy compressors are supposed to be transparent.


Make it short: no, i can not tell the difference, although i should since i used very high compression rate (10x). Have you listened to the samples?

Can not tell difference between these samples

Reply #3
So if I understand correctly, you're saying that you can't hear the difference between lossy and lossless files? My take is that it means that ogg is doing its job. Good lossy compressors are supposed to be transparent.


Make it short: no, i can not tell the difference, although i should since i used very high compression rate (10x).


Why does that mean you should be able to hear a difference?

Can not tell difference between these samples

Reply #4
Why does that mean you should be able to hear a difference?

You're pulling my leg, right?
The basis of lossy formats is cutting out some frequencies. Since i used high compression rate when compressing to ogg (lossy format), many of them must have been cut out and that should be strongly noticeable in the quality of sample. As i said, i'd accept that i'm deaf to those frequencies if it wasn't for soundexpert tests, where i had no problem distinguishing between wav (lossless) and anything below 320+ kbit/s in lossy formats.
That is why i asked can you tell the difference between the samples i pasted, since not being able to tell quality difference between original and 10x compressed format can not be right.

Can not tell difference between these samples

Reply #5
Why does that mean you should be able to hear a difference?

You're pulling my leg, right?


No I'm not.  Why do you think you should be able to tell a difference?

The basis of lossy formats is cutting out some frequencies.


No its not.

As i said, i'd accept that i'm deaf to those frequencies if it wasn't for soundexpert tests, where i had no problem distinguishing between wav (lossless) and anything below 320+ kbit/s in lossy formats.


If you want to test if you're deaf, I recommend a hearing test, not Soundexpert (which is incidentally pretty useless). 

That is why i asked can you tell the difference between the samples i pasted, since not being able to tell quality difference between original and 10x compressed format can not be right.


If everyone else could tell the difference would that change the fact that you could not?  If no one else could, would that change the fact that you couldn't?  What does that even tell you?

Can not tell difference between these samples

Reply #6
Could we not show OP just a little bit of love? Especially since the topic
' "Audiophile" listening event @ Definitive Audio in Seattle, Atkinson to demonstrate "evils of MP3" '
shows how much misleading propaganda is out there?

Could one not just say that good modern encoders like vorbis are designed to throw away only those parts of the signal that isn't registered, anyway, and that not being able to hear a difference between lossless and a modern lossy encode at some reasonable bit rate is perfectly normal, and not a sign of any kind of deafness?

You know, without all the aggro and n00b-bashing?

Can not tell difference between these samples

Reply #7
The Soundexpert site is using a special technique to "amplify" the effects and artifacts of lossy encoding.  That is why you hear a difference on their site, but not in your own material.

You are unlikely to easily hear differences between the original file and Ogg Vorbis at 141kbps ("10x compression")

Can not tell difference between these samples

Reply #8
The basis of lossy formats is cutting out some frequencies.


Not exactly. The basis of lossy formats is cutting out some frequencies that are difficult or improssible to hear under the current dynamic conditions in the music.

Quote
Since i used high compression rate when compressing to ogg (lossy format), many of them must have been cut out and that should be strongly noticeable in the quality of sample.


As others have already and correctly pointed out, the fact that some frequencies were removed is not a reliable indicatior that there should be an audible difference.  A game is being played and if played well, you never notice that the game is being played.


Quote
As i said, i'd accept that i'm deaf to those frequencies if it wasn't for soundexpert tests, where i had no problem distinguishing between wav (lossless) and anything below 320+ kbit/s in lossy formats.


But, the Soundexpert tests were biased to make the difference more audible than it might be in general.

Quote
That is why i asked can you tell the difference between the samples i pasted, since not being able to tell quality difference between original and 10x compressed format can not be right.


If 10x reduction is done right and the music is not too challenging to code, then the expected outcome is that you will not easily hear a difference. You may not be able to hear a diffeence no matter how hard you try.

Lossy coding is based on the idea that music is listened to by humans and that in gereral, human listeners have certain predictable failings.

Can not tell difference between these samples

Reply #9
I heard the two samples and I also can't find a real difference. But the samples aren't  good for testing the codec.

The example "hope"  contain no  frequencies over 16Hz and the gain of frequencies over 8KHz is under -75dB.

The example "freude"  contain no frequencies over  10,5kHz and the gain of frequencies over 8 KHz is under -80dB.

Higher frequencies are often cut  by lossy codecs at low bitrates, but your samples have no high frequencies. Maby it would be easier to hear a difference with other samples.


Can not tell difference between these samples

Reply #10
The basis of lossy formats is cutting out some frequencies.

This is misunderstood. Information is removed, frequencies generally are not. The basic concept is to replace high resolution audio with low resolution in areas of frequency at time where the encoder predicts that you will not be able to hear the difference.

Can not tell difference between these samples

Reply #11
The basis of lossy formats is cutting out some frequencies.

This is misunderstood. Information is removed, frequencies generally are not. The basic concept is to replace high resolution audio with low resolution in areas of frequency at time where the encoder predicts that you will not be able to hear the difference.


In many cases signals that are composed of discrete frequencies are replaced with bands of noise. Thus the spectral balance of the signal is preserved, but  much of the information is removed.

Can not tell difference between these samples

Reply #12
Arnold, resolution is reduced on a per-band basis.  If you wish to characterize it as replacing signal with noise, so be it, though that doesn't really go very far in explaining how lossy encoding works.  On the contrary, it could give someone the impression that you don't really know how it works.

Can not tell difference between these samples

Reply #13
Arnold, resolution is reduced on a per-band basis.  If you wish to characterize it as replacing signal with noise, so be it, though that doesn't really go very far in explaining how lossy encoding works.  On the contrary, it could give someone the impression that you don't really know how it works.


It seems to me like this is a misunderstanding over wording. You say "resolution is reduced on a per-band basis" I say that "discrete signals are replaced with bands of noise."  I don't see a critcial difference.

Q: What happens when you reduce just the resolution of a signal per Shannon?

A: The signal gets noisier

right?

Can not tell difference between these samples

Reply #14
Would this be another way to describe this? The contents of a frequency band are replaced with a less-accurate version, which is, in effect, a noisier version?

Can not tell difference between these samples

Reply #15
Benski: yes i know about the amplification in soundexpert, that makes their testing set up awkward (maybe even biased, as some have mentioned).

To be precise, i used the compression level 1 for the samples, which produces 80kbps rate - i wouldn't have really minded that much if wasn't able to distinguish from 160kbps onwards, but this seemed just odd to me (taking nothing away from the genius of the developers of vobris).

Musicreo: thanx for the advice, it didn't come to my mind to observe the frequencies spectra of samples i use. Surely you used a windows OS tool for that? I've also spent some time looking for proper samples for testing, but didn't come across any substantial set so far. One ABX tool for linux i tried so far also failed me.

Regarding the quick note i dropped on the vobris algorithm itself, i definitely do not know how exactly it works, i actually made a presumption from image processing course i'm doing now - quite often the basic approach there is to get rid of fine details to some extent (high frequencies from Fourier domain) while letting the low frequencies pass through. Incidentally Discrete Cosine Transform (used in Vobris) will be one of the topics in the exam next week. Another funny incident is that my laptops backlight got busted yesterday, so finding out how to replace it is a bigger headache for me than bitrates right now.

Thx for replies.
Peace

Can not tell difference between these samples

Reply #16
Would this be another way to describe this? The contents of a frequency band are replaced with a less-accurate version, which is, in effect, a noisier version?


Non-critical coefficients of the MDCT are replaced with quantized versions that can be stored with less bits and that are "close enough".  Many (most) are quantized to zero or one even though the true value may lie somewhere in between.  Since a zero coefficient can be stored with a single bit and a one can be stored with 2 bits, this is where most of the data reduction comes from.  Distortion is probably a more accurate term than noise (noise can imply broad-band "white noise" to most people).

Can not tell difference between these samples

Reply #17
It seems to me like this is a misunderstanding over wording. You say "resolution is reduced on a per-band basis" I say that "discrete signals are replaced with bands of noise."  I don't see a critcial difference.

Q: What happens when you reduce just the resolution of a signal per Shannon?

A: The signal gets noisier

right?

The difference is that the signal is still there;* it is not "replaced" with noise; it is quantized. Also it is quantization distortion, not noise. In some circumstances quantization distortion can be modeled as signal + noise, but those circumstances don't always hold in an encoder.

*Except in the case where a band is quantized to 0 bits. This does happen, especially at lower bit rates.

Can not tell difference between these samples

Reply #18
Checking back in on the topic:

After compressing my music library to ogg, uploading it on my Cowon IAudio 7 and going with it on the month long journey (meaning not having possibility to listen to other formats at the time), the artefacts i mentioned in my first post, to which i didn't pay much attention during the testing, eventually drove me mad. They popped pretty much on everything i listened to and as much as i love music, i ended up listening to music with clenched teeth. The artefacts are fairly easy to recognise on e.g. the base line at the beginning of Dream Theaters - In the presence of enemies Pt. 2 (ogg level 10 sample), it's an annoying twittering sound echoing base guitar, which is really escalated when i'm listening with sennheiser eh350, still noticeable with cheap in-ear headphones i tried. Changing the encoding level (encoding with lame using foobar) does not change anything. I can not hear the artefacts while listening through my computer (with headphones) though. Noticed no such artefacts while listening to mp3 version on my player as well.

So, while being an open source patent-free coding supporter and not having much of an lossy alternative for my iaudio 7, i'll be moving back to flac.

It would be interesting to hear other opinions on the sample though, especially from owners of cowon players, i'm kind of amused to know the root of the problem.

Pys

Can not tell difference between these samples

Reply #19
So if I understand correctly, you're saying that you can't hear the difference between lossy and lossless files? My take is that it means that ogg is doing its job. Good lossy compressors are supposed to be transparent.


Make it short: no, i can not tell the difference, although i should since i used very high compression rate (10x). Have you listened to the samples?


I would not call 10x lossy compression "very high".

Can not tell difference between these samples

Reply #20
So if I understand correctly, you're saying that you can't hear the difference between lossy and lossless files? My take is that it means that ogg is doing its job. Good lossy compressors are supposed to be transparent.


Make it short: no, i can not tell the difference, although i should since i used very high compression rate (10x). Have you listened to the samples?


I would not call 10x lossy compression "very high".


it does not really matter now any more, at least not for me and my player  i don't mind listening to compressed files any more (not least as a computer science orientated student, understanding the trade-off of data compression), it's just that this particular format (particularly on my player?) and the etching sound that comes with it is unbearable.

Can not tell difference between these samples

Reply #21
Your -q 10 Vorbis sample is fine. I decoded it with foobar and created a time aligned reference sample from my lossless "Dream Theater - In the Presence of Enemies Part 2" rip. I couldn't hear a difference in a quick ABX test. In my opinion the sample is not very demanding. I guess -q 4 or even a lower setting would be fine for this part of the song.

Probably your player is causing the problem.

Topic: Cowon iAudio 7 (fw 1.17) fails on Vorbis, "Electrical" noise: http://www.hydrogenaudio.org/forums/index....showtopic=66999

Topic: iAUDIO 7 Vorbis distortion: http://www.jetaudio.com/forums/showthread.php?t=13253

From a Cowon iAudio 7 review at http://www.linuxjournal.com/magazine/cowon-iaudio-7 , comments by "fresta" and "pellgarlic":
Quote
Warning: ogg support broken

Submitted by fresta (not verified) on Wed, 04/01/2009 - 06:03.

Just a word of warning: ogg support is seriously broken on the IAudio 7 player. Most importantly, there is a really annoying noise in the treble when playing some ogg files with bass tones around 100 kHz, see this post at cowonamerica.com. For me it spoils the listening experience completely for about every fourth ogg file in my collection.

Cowon has been aware of it for almost two years now, and done nothing. It's actually a very simple bug and has been fixed on the D2 I think, though several other models from Cowon still suffer from it.

Also, ogg tags parsing is broken in music mode so it is not possible to browse ogg files by artist, album or genre etc. You have to use folder mode.

The IAudio7 definitely has one of the best hardware on the market, the engineers at Cowon has done a good job there. Pity that it should be spoiled by incompetent software programming and lousy support.

Quote
vorbis distortion issue fixed

Submitted by pellgarlic (not verified) on Fri, 05/29/2009 - 09:22.

hi, i just wanted to share some new info with anyone who already owns, or is thinking about getting an iaudio 7 - there is a new firmware release - 1.18 (go here: http://www.cowonglobal.com/ to get it) - which fixes this particular problem (the vorbis distortion). it was a significant problem for me (as my music collection is predominantly in ogg-vorbis format), but i didnt find out until after i had purchased the player. to be honest, i couldn't find another player which ticked all the boxes for me that the i7 does, so i was pretty peeved, and loathe to start the hunt for an alternative, so i emailed cowon support (as many other people have probably done too) and got a few replies, the last one being a few days ago, saying they were releasing a fix  i have now downloaded it and tested it, and can confirm that it has resolved the vorbis-distortion-caused-by-low-frequency problem. may i say, on a personal note... "woo-hoo!"

Can not tell difference between these samples

Reply #22
Hi, I'm new to the forum but have been recording and listening to music for many years. The problem with all these tests is that they lack a truly analog reference. Real sound happens in the analog domain. To truly be able to determine the differences made by the various encoding schemes (including CDA and WAV), you need an analog reference.

I recently purchased a new HTC phone and wanted to put some music on it. I read through the posts on this forum and tested several different lossy formats and compared them to reference analog recordings. There were differences but I also discovered that the biggest difference wasn't caused by the lossy format but by the electronics in the phone.

When I played back the ripped material on the reference studio system they sounded great for 192 VBR compressed audio but the electronics in the phone degraded this down a notch.

I'm writing this to express that lossy compression has its place in the world and that its use is relative to the desired objective. If you want the finest recordings and have the system capable of reproducing them then you should forget about portable audio, car audio, etc. If, however, you love music, and the recording quality needs only be good, then anything above 128 VBR can give you endless hours of portable music bliss.  :-)

I would also like to thank the contributors and admins of this forum for the great Wiki and the excellent information provided. This forum saved me endless hours of trial and error ripping. Great job!  :-)


Can not tell difference between these samples

Reply #23
Hi, I'm new to the forum but have been recording and listening to music for many years. The problem with all these tests is that they lack a truly analog reference. Real sound happens in the analog domain. To truly be able to determine the differences made by the various encoding schemes (including CDA and WAV), you need an analog reference.


Actually, hearing is digital thanks to the way neurons work.  Reality is digital too, since energy states are quantized. So the reference should be a digital recording.

(Yes, these are silly arguments, but no sillier than yours.)

The ultimate reference for any recording/playback technology, is the sound of the original event.  Since most of us aren't in the room as the recording is happening, we don't have access to that.  And even if we were there, chances are the original 'event' was at least partially lined-in directly to the board, with the 'ambience' being added electronically or digitally later. 

For recordings we want to encode into another medium/format --- which is what most of us actually have access to -- the reference for determining difference is the recording before encoding.  Nothing more or less.  It's doesn't have to be analog.

Btw, tape and vinyl are a sort of 'encoding scheme' too. (And they're lossy.)

 

Can not tell difference between these samples

Reply #24
It is technically more difficult to do an ABX with an analog reference but it can ans has been done. The results indicate that some digital equipment is not up to par and a lot of analog equipment introduces distortion but Niquist and Shannon were right - digital representations have quantization and bandwidth limitations but are otherwise true to the original.

Once you find the point where you cannot ABX analog vs. digital, it is fair to start doing digital vs. digital comparisons.