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: Lossy Codecs Frequency Evaluation and a question (Read 8065 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Lossy Codecs Frequency Evaluation and a question

Spectral Frequency Display Screenshots (using Adobe Audition v.2.0)
The song is "Cross My Heart and Hope to Die" from the Sentenced album "The Cold White Light". It is a melodic/atmospheric metal song with heavy guitars that produce a "wall of sound" throughout the song. Quite demanding song that makes VBR presets generate larger than usual files. 
Extraction and encoding done with Easy CD-DA Extractor v.9.0.1 (Build 2)

Lossless (.wav (Standard) & .wv (Wavpack))


Lossless (.wav & .wv) to Nero Audiophile Preset [Nero AAC Codec 3.2.0.15]: 290 kbps, 8.57 MB


Nero Audiophile Preset [Nero AAC Codec 3.2.0.15]: 290 kbps, 8.57 MB


Nero Audiophile Preset [Nero AAC Codec 3.2.0.15] Second Time: 290 kbps, 8.57 MB


Nero Extreme Preset [Nero AAC Codec 3.2.0.15]: 250 kbps, 7.38 MB


Ogg Vorbis 1.1 VBR Quality 9 (320 kbit/s) [Xiph.Org libVorbis I 20050304]: ~320 kbps, 9.43 MB


Musepack Braindead Preset [integrated]: ~259kbps, 7.61 MB


MP3 CBR 320 kbit/s [Lame 3.97]: 320 kbps, 9.40 MB


Although there are always losses in frequencies when moving from lossless to lossy, it is also interesting to see how much each codec is altering the original sound picture. If you observe somewhere in the mid-frequencies (e.g. the "quiet break" between 2:30 and 2:56), Nero AAC excels in that respect producing an almost identical frequency display and is followed by mpc and then come ogg and mp3.
Actually, both Nero AAC presets seem to outdo the competition by displaying the less frequency loss and picture alteration.

What troubles me is the slightly different display of "lossless to Nero AAC" and "Nero AAC extracting directly from the CD". I have no idea why is that and which should be the preferred method.
If someone can enlighten me on this, please do.
Any general comments also welcome.

Lossy Codecs Frequency Evaluation and a question

Reply #1 –
A search of this board will reveal that the consensus is that to evaluate sound that the ears are better than the eyes.  Nice pictures, though. 
Nov schmoz kapop.
 

Lossy Codecs Frequency Evaluation and a question

Reply #2 –
General comment: It shouldn't matter what it looks like at all, just how it sounds

Kristian

Lossy Codecs Frequency Evaluation and a question

Reply #3 –
Quote
What troubles me is the slightly different display of "lossless to Nero AAC" and "Nero AAC extracting directly from the CD". I have no idea why is that and which should be the preferred method.
If someone can enlighten me on this, please do.
[a href="index.php?act=findpost&pid=372920"][{POST_SNAPBACK}][/a]


I'm quite sure the .mp4 *file encoded* from wav/wv to nero versus the one encoded from CD isn't different.
You have just found out why these types of graphics should not be taken seriously when comparing. Even more if it is a zoom of the whole song.

Basically, the program is showing you a zoomed out image of what it has, and the differences come from this zooming out. Of course, there could be the strange fact that both files had a different length (by some milliseconds or so). That would change the graphic aswell.

Lossy Codecs Frequency Evaluation and a question

Reply #4 –
If you want your music to *look nice*, use a lossless codec - lossy codecs aren't meant for that.
I am arrogant and I can afford it because I deliver.

Lossy Codecs Frequency Evaluation and a question

Reply #5 –
The test was carried out with an audiophile's point of view in mind: what codec produces results that are most close to the original.

These results (nero aac being more close to the original-mostly missing some high frequencies, ogg including higher frequencies than mpc etc.) are observed monotonously with any file I test, so I think it can be safely said the "Spectral frequency display" is representative of how a codec behaves and what changes it makes to the original.

I'm under the impression that you guys underestimate the usefulness of such graphical representations.

Personally, I find information like that more reliable than any "heed closely and spot the difference " listening test when we are talking about audiophile quality levels.

I'm not prefessing to have delivered the ultimate truth here but I just can't ignore information like that when choosing lossy formats for high-fidelity.
Do you think Nero AAC just stands lucky and produces such close to lossless graphical displays whereas the others are just treated unjustly by the sound analysis software?Or do you profess that human perception is better than sound analysis for detecting differences between sound files at these levels of compression?Could it be possible to give ogg or mp3 the credit for producing the closest to the original compressed audio, when they have distorted the frequency and waveform display so much?

About my question, personally it's a little difficult for me to go on and blame Adobe Audition for the anomaly because Audition proved to be quite consistent and accurate.I mean the .wav and .wv graphics were 100% identical and so were the ones of the nero aac files produced out of them.
The two nero acc files which were produced without conversion were also identical with each other.
It just so happens that the two nero aac files made by conversion and the two nero aac files made on the fly have different sound forms , a fact that is represented in both Waveform and Spectral Frequency displays.They differ very slightly but they do.
All files have the same duration but I don't know about milliseconds.Even if ,strangely enough, there are spare milliseconds, would that justify this kind of changes in the displays?
I honestly don't know what to think.I just hope some "expert" makes sense out of this.

Lossy Codecs Frequency Evaluation and a question

Reply #6 –
I vote either "Troll" or "a bit slow in understanding and following advice".

Someone else already adviced you to search the forum for why waveform-difference analysis or spectral analysis has not much to do with soundquality.

FYI, when you registered to this forum you agreed to this rule:

Quote
8. All members that put forth a statement concerning subjective sound quality, must -- to the best of their ability -- provide objective support for their claims. Acceptable means of support are double blind listening tests (ABX or ABC/HR) demonstrating that the member can discern a difference perceptually, together with a test sample to allow others to reproduce their findings. Graphs, non-blind listening tests, waveform difference comparisons, and so on, are not acceptable means of providing support.


If you want "audiophile"-style discussions, then you're on the wrong board.


- Lyx
I am arrogant and I can afford it because I deliver.

Lossy Codecs Frequency Evaluation and a question

Reply #7 –
OK, one test is worth a thousand opinions.

Regardless of what they look like, the way to distinguish sounds is with your ears not your eyes.  That is why listening tests are used to compare codecs.  It all kind of makes sense, doesn't it??  Unless you are in the marketing department where pretty pictures make sense.  YMMV. 
Nov schmoz kapop.

Lossy Codecs Frequency Evaluation and a question

Reply #8 –
I'd like to tell something about this.

Someone not used to scientific evaluation of audio encoding would immediately place the interventions of Olyrous here on the "objectivist" side : measures and facts, while other people would be obviously on the "subjectivist" side : refusing any measure or any number as "irrelevant", repeating that what you hear is what you hear, no matter what science have to say, and "I don't have to listen to others advices because only what I here is important, so just stop ranting about numbers and graphs, music is not graphs". That's the definition of subjectivism. And that is exactly the answer that I got from ultra-subjectivist forums when talking about blind listening tests, just before being banned for discussing audio quality evaluation (it was not in Steve Hoffman's forum).

Asking for the newcomer to perform a serch won't be of any use. Very difficult to choose the right criterions when we don't know what we are looking for. Spectrum analysis ? Frequency losses ? These won't lead to the "why graphs are irrelevant for audio quality evaluation" threads.

So to make it short, hydrogenaudio is an objectivist community, not a subjectivist one.
But the matter of psychoacoustic audio compression is very special. Actually, the fact that an encoder produces a file different of the original can be a sign of good quality, that's why we ask to the encoder : to remove as many frequencies and sounds as possible without audible change.
For similar audible quality, the codec that compresses the most, i.e. that produces the files that is the most different from the original, will be the best !

Psychoacoustic models are there in order to get frequencies removed (in order to compress the filesize) in an inaudible way. That's why spectrum analysis, performed outside a psychoacoustic model, cannot give any useful information about sound quality. Are the missing frequencies the ones that ought to be removed or the ones that should not be removed ? Sonograms can't answer this crucial question.

That is why the only way to get an idea of the sound quality of a codec is a listening test run under controlled conditions. Protocols ensuring that no bias affects the listening test exist, and they are compulsory in Hydrogenaudio.
The most basic one is called ABX, see http://wiki.hydrogenaudio.org/index.php?title=ABX
Some protocols more useful are often used, like ABC/HR.

The way to get relevant information from these listening test is to make them "double blind", and to get their "statistical significance". this is explained here : http://www.hydrogenaudio.org/forums/index....howtopic=16295&

You'll find some more links in the FAQ : http://www.hydrogenaudio.org/forums/index....t=ST&f=5&t=7516

Lossy Codecs Frequency Evaluation and a question

Reply #9 –
Quote
Although there are always losses in frequencies when moving from lossless to lossy, it is also interesting to see how much each codec is altering the original sound picture.

actually no it is not interesting. as long as the sound match the original for my ears i could care less about how the compressed file "looks" in audition.
Be healthy, be kind, grow rich and prosper

Lossy Codecs Frequency Evaluation and a question

Reply #10 –
I forgot to clarify one thing : all the above posts are in fact "objectivist".

When people here said that "only what we hear matters" they meant "in controlled double blind listening tests". Not listening to oneself's preferences.

Lossy Codecs Frequency Evaluation and a question

Reply #11 –
Restating what Pio2001 is saying: It's quite possible to produce an encoder that looks better on the spectrogram than another encoder, but sounds objectively worse. Compression artifacts have nothing to do with the frequency response.

Another way to put it: If you optimize purely for an optimal spectral response, you will degrade the sound quality. Period. You are taking away bandwidth that could be used for more useful things and applying it to a problem that does not affect sound quality as much. At higher bitrates this is somewhat less of a concern but the exact same issue still applies.

Lossy Codecs Frequency Evaluation and a question

Reply #12 –
Quote
Although there are always losses in frequencies when moving from lossless to lossy, it is also interesting to see how much each codec is altering the original sound picture. If you observe somewhere in the mid-frequencies (e.g. the "quiet break" between 2:30 and 2:56), Nero AAC excels in that respect producing an almost identical frequency display and is followed by mpc and then come ogg and mp3.
Actually, both Nero AAC presets seem to outdo the competition by displaying the less frequency loss and picture alteration.
[a href="index.php?act=findpost&pid=372920"][{POST_SNAPBACK}][/a]


Hello everyone. I finally have the proof that JPEG is better than GIF and PNG. I have attached audio files compressed with all these image codecs. From these audio files, it's clear that JPEG sounds *much* better. You can clearly hear how JPEG preserves the bass and stereo coupling better. On the other hand, we can hear a bit of pre-echo, which is not present in GIF. I think listening to images is a lot better than merely looking at them because there are a lot more things you can here that you don't necessarily see. 

Lossy Codecs Frequency Evaluation and a question

Reply #13 –
Quote
Hello everyone. I finally have the proof that JPEG is better than GIF and PNG. I have attached audio files compressed with all these image codecs. From these audio files, it's clear that JPEG sounds *much* better. You can clearly hear how JPEG preserves the bass and stereo coupling better. On the other hand, we can hear a bit of pre-echo, which is not present in GIF. I think listening to images is a lot better than merely looking at them because there are a lot more things you can here that you don't necessarily see. 
[{POST_SNAPBACK}][/a]

Oh no. It's my first every hydrogenaudio [a href="http://www.hydrogenaudio.org/forums/index.php?showtopic=14103]post ("New Lossy Audio Codec")[/url] come back to haunt us.

To the original poster - the point that we are making is that lossy audio codecs strive to make a file sound as similar to the original as possible for a given bitrate (or be as small as possible for a given quality degradation). Measures that are useful for other audio components (THD, SNR, frequency response, etc) are not generally useful for comparing lossy audio codecs.

Lossy Codecs Frequency Evaluation and a question

Reply #14 –
Quote
I'm under the impression that you guys underestimate the usefulness of such graphical representations.

I agree that this analisys is not useful when you want to compress music that will be listened.
But this analysis can maybe useful when you want to compress some audio data that will be not heared by a human, for example some transmission channel that uses sound as a transport. In this case an objective analisys is correct.
Howerver would be nice to see also the graphs of the latest aoTuV version.

Lossy Codecs Frequency Evaluation and a question

Reply #15 –
I recall reading another thread in which a statement was made that the person responsible for the development of the Blade mp3 encoder did it largely relying on spectral graphs.

Where is Blade today? 

Quote
I agree that this analisys is not useful when you want to compress music that will be listened.


For the purpose of comparing audio codecs that compress music, why will we want to compress and then compare audio signals that are out of human hearing range? After all, isn't the crux of psychoacoustic compression about reducing file sizes by removing as many inaudible (masked) signals as possible (of course while keeping the remainder as close to the original as possible)?