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: MPC vs OGG (Read 30250 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

MPC vs OGG

Reply #25 –
I try to use EAQUAL to adjust some properties of the codec which do not affect temporal structure. This is helpful in debugging some parts of psychoacoustic model.

MPC vs OGG

Reply #26 –
Quote
I try to use EAQUAL to adjust some properties of the codec which do not affect temporal structure. This is helpful in debugging some parts of psychoacoustic model.
So then these results show that ogg is 18.5 times better than mpc on this sample in "parts of psychoacoustic model" which "do not affect temporal structure". I sure hope Alex will implement the advanced model sometime.

MPC vs OGG

Reply #27 –
Quote
Originally posted by layer3maniac
I'm not "trying to prove" anything. I'm just posting results. It's funny that you mentioned that though. That's EXACTLY the reason I don't trust Joe Blow's abx results. Too many people ARE "trying to prove" something instead of letting the chips fall where they may. An abx test isn't blind at ALL. You KNOW which codec you are testing as B in any given test. By the way, I'm curious why (and how) you use eaqual to tune your encoder considering you say you don't trust the conclusions it gives...


This is a good argument, but it also applies to your results. You may be subconciously biased. I would be happy to do a blind test, especially as our results contradict.

--
GCP

MPC vs OGG

Reply #28 –
Quote
By the way, I'm curious why (and how) you use eaqual to tune your encoder considering you say you don't trust the conclusions it gives...


Me too...

MPC vs OGG

Reply #29 –
Quote
Originally posted by layer3maniac
So then these results show that ogg is 18.5 times better than mpc on this sample in "parts of psychoacoustic model" which "do not affect temporal structure". I sure hope Alex will implement the advanced model sometime.
It certainly doesn't mean that it's 18.5 times or in fact any better than MPC. All it does show is that EAQUAL's psymodel thinks it's better. Even Alexander says EAQUAL's psymodel isn't very sophisticated and is few years old, much older than MPC's or Vorbis' model.
Now you are measuring the quality of other codecs with newer psymodels with not so sophisticated older model and even believe something is 18.5 times better (how can you possibly translate that to audible quality anyway).. Only thing you can conclude is that EAQUAL may like VORBIS' model a bit more, but is that a proof that the quality is truely better? Certainly not.
http://www.hydrogenaudio.org/forums/showth...s=&threadid=714

Point is, if you can't judge codecs yourself, don't take the "newbie" way and try to think there's an easy way to judge quality...
Juha Laaksonheimo

MPC vs OGG

Reply #30 –
Quote
Originally posted by layer3maniac
By the way, I'm curious why (and how) you use eaqual to tune your encoder considering you say you don't trust the conclusions it gives...
I got the impression Ivan uses it only for low bitrate tuning, by going trough lots of settings and verifying results afterwards by himself. I would think EAQUAL's (imperfect) model is more useful, the larger the differencies are like it's the case at low bitrate. But to judge high bitrate state of the art models (Vorbis,MPC,Psytel 2.0) with few years old "ordinary" psychoacoustics and to trust the results would be simply stupid.
Juha Laaksonheimo

MPC vs OGG

Reply #31 –
JohnV was right - I'm using it on very low bitrate only, I am quite sure that high bitrates are tuned well

MPC vs OGG

Reply #32 –
The nice thing with audio quality evaluation is, there is and will be never an objective proof of quality. Neither a big listening test nor an objective measurement tool, no matter if good or bad, can yield to an objective proof of quality, because every proof of quality can only be *subjective*. Everyone has the right to say about any rating, listening test or measurement, that this doesn't reflect his opinion.

What I want to say, if you are of opinion that one special codec is best, then this codec _is_ best (for you), whatever any listening tests or measurement methods would say.

Best regards,
Alexander

MPC vs OGG

Reply #33 –
Quote
I got the impression Ivan uses it only for low bitrate tuning, by going trough lots of settings and verifying results afterwards by himself.
And I went the other way around. The 195 kbps ogg file sounded BETTER than the 231 mpc insane file so I verified the results afterward. The difference is? I STILL don't understand though. If Ivan can't trust the results due to eaqual's "imperfect model", why does he use it at ALL? You don't see the dichodemy in this? Either it's useful or it's not.
Quote
I would think EAQUAL's (imperfect) model is more useful, the larger the differencies are like it's the case at low bitrate.
Larger differences than EIGHTEEN AND A HALF TIMES???
Quote
But to judge high bitrate state of the art models (Vorbis,MPC,Psytel 2.0) with few years old "ordinary" psychoacoustics and to trust the results would be simply stupid.
I think judging codecs by using abx results from people with individual biases, preferences, priorities, and hearing loss is EVEN STUPIDER! At least with a tool like this we get completly unbiased results. But hey, if you don't like the results due to your OWN biases, preferences, priorities, and hearing loss then you can just blame the tool.

MPC vs OGG

Reply #34 –
Quote
Originally posted by JohnV
I would think EAQUAL's (imperfect) model is more useful, the larger the differencies are like it's the case at low bitrate. But to judge high bitrate state of the art models (Vorbis,MPC,Psytel 2.0) with few years old "ordinary" psychoacoustics and to trust the results would be simply stupid.


ITU-R BS.1387 was not designed for low or middle quality, but the term middle quality of course changed over the years.
I did some research about psychomodels and all I found was that there has not really much changed in the last decade and nearly all psychomodels are based on approaches that are older than one decade.
I talk about the algorithm and not its tuning. And I don't think that this must be necessarily negative.

I don't want to say that EAQUAL is great. I know the limitations of the algorithm and I found many things in mind that could be done better. You only have to know in which cases you can use it successfully and in which cases not.

Don't trust anyone. Just listen!

Alexander

MPC vs OGG

Reply #35 –
Uhh.. huh... well - I'll try to explain

- First of all, it is very hard to judge state-of-the art codecs at very high bitrates because they are transparent for many people, and the confidence of the listening test will start to decrease because many subjects will try to make up some "invisible" artifacts.


- At the lower bitrates, where distortion is impossible to avoid (like 96 kbps for 44.1 kHz stereo) it is still possible to make that distortion sound "more pleasing" - take a look at Liquid Audio (fhg aac) - files encoded at 96 kbps are distorted, but in a least perceptually offensive way.

- I think that at -extreme mode of psytel AAC there is no NMR distortion at all,  hence I think that it is perceptually transparent. If someone is going to prove me wrong, I would like to hear the samples. Same goes to Ogg -q <something, maybe 6 or 7, or 8) and MPC -extreme. Trying to measure "how good" a transparent codec is (if it is transparent) is a little bit complicated, and that certainly can't be done with tool that uses perceptual model with known shortcomings...

Going around saying that one transparent (or near transparent) codec is 18 times better than another transparent codec is a nonsense - like saying that one infinity is 18 times bigger than another infinity (ok... don't start linear algebra here, please

MPC vs OGG

Reply #36 –
Quote
First of all, it is very hard to judge state-of-the art codecs at very high bitrates because they are transparent for many people, and the confidence of the listening test will start to decrease because many subjects will try to make up some "invisible" artifacts.
I couldn't agree MORE.
Quote
Going around saying that one transparent (or near transparent) codec is 18 times better than another transparent codec is a nonsense - like saying that one infinity is 18 times bigger than another infinity (ok... don't start linear algebra here, please
First, there is NO SUCH THING as a transparent lossy codec. That's like calling a mature 11 yr old an adult. Second, I NEVER SAID that ogg is 18 times better than mpc. I said that ON THIS SAMPLE, THIS TOOL says that ogg is 18.5 times better than mpc. And that's TRUE, like it or not... I have another question though. If you messed with some settings in aacenc and ran an eaqual test and it said that a sample encoded using the new settings was 18.5 times better wouldn't you take a closer look at it?

MPC vs OGG

Reply #37 –
>I said that ON THIS SAMPLE, THIS TOOL says that ogg is 18.5 times better than mpc. And that's TRUE, like it or not... I have another question though. If you messed with some settings in aacenc and ran an eaqual test and it said that a sample encoded using the new settings was 18.5 times better wouldn't you take a closer look at it?

I guess I would but I would also note that you can hardly call a 10second sample statistically meaningful...

(Up next: the world's lousiest reasoning - blame it on work stress )
I also don't think one can adhere too much credibility to the rating system of EAQUAL since if this was the best way to judge audio quality, it's psymodel should be used to create the best lossy encoder then.
No inspiration

MPC vs OGG

Reply #38 –
Quote
I did some research about psychomodels and all I found was that there has not really much changed in the last decade and nearly all psychomodels are based on approaches that are older than one decade.
Apparently JohnV knows VERY little about the history of psychomodels.

MPC vs OGG

Reply #39 –
Quote
Originally posted by layer3maniac Larger differences than EIGHTEEN AND A HALF TIMES???
What 18.5 times? What does it mean exactly regarding audible quality? Yes, there are certainly larger AUDIBLE differencies at low bitrates. That 18.5 is just a value from EAQUAL's model.

Quote
I think judging codecs by using abx results from people with individual biases, preferences, priorities, and hearing loss is EVEN STUPIDER!
Hearing loss I could understand, but since ABX is blind test (you cannot know what the X-sample is) it will objectively show if the tester can distiguish A from B. Bias or individual preferences  has nothing to do with that.

What you don't seem to get is that since EAQUAL is just another implementation of another (older imperfect) "ordinary" psychoacoustic model, there's no guarantees it's better than experienced listeners. In fact, I would trust experienced listerer 99.9% more than EAQUAL result.

I would love to see ITU-R BS.1387 model implemeted as closely as possible in some lossy codec. I'll bet it wouldn't be better than Vorbis' or MPC's current model (that's 100% certain as far as temporal accuracy goes). Where's the proof ITU-R BS.1387 is better model than others? As far as I can see and hear, it's not better, it's clearly worse. Why would anybody take its results as a fact??
Quote
At least with a tool like this we get completly unbiased results.
Unbiased meaning here only psycological effect in non-blind test. This model could be regarded even more biased than a human doing abx, because it can give consistantly bad results (it's biased towards certain response of its model). Human tester may be better or worse even depending on time of the day.
Juha Laaksonheimo

MPC vs OGG

Reply #40 –
Quote
Originally posted by layer3maniac
Apparently JohnV knows VERY little about the history of psychomodels.


It is true that basic psychmodel structure hasn't changed for a decade but advanced codecs do have "fine tweaking" of various parameters in order to achieve best NMR on a specific coding preset.

For example, take a look at ISO 11172-3 and 13818-7 so-called "Psychoacoustic model 2", this is a model recommended by ISO for MP3 and AAC encoders. Many encoders are based on this.

However, the author who wrote this part of the ISO doc based it on the model that his company uses, but with lots of simplifications. Parameters used in are actually took from one particular codec preset and configuration, other tools are also simplifications of the optimized psychoacoustic model they use in their products - they continued to tweak the model for all cases long after the ISO 11172-3 and 13818-7 standards were finished.

MPC vs OGG

Reply #41 –
Quote
Hearing loss I could understand, but since ABX is blind test (you cannot know what the X-sample is) it will objectively show if the tester can distiguish A from B. Bias or individual preferences  has nothing to do with that.
In what alternate universe is an abx test blind? You KNOW what codec B was made with. That's NOT blind. If you're biased toward the psymodel codec B used, the results will reflect that. Some people are extremely sensitive to pre echo, some to dropouts, some to post echo, some to high or even low frequencies, some, like beatles, to nuances of overall depth or warmth, and some to stereo separation. ABX tests are skewed by ALL these biases, preferences, priorities and hearing difficulties.

MPC vs OGG

Reply #42 –
Quote
Originally posted by layer3maniac
Apparently JohnV knows VERY little about the history of psychomodels.
Heh . As I was saying, EAQUAL is based on "ordinary" psychoacoustics. Newer codecs has the same "ordinary" psychoacoustics with added or improved options like MPC's CVD, non-linear spreading functions, better tonality estimation algorithms. The key point is the tweaking.

You are trusting blindly to a psymodel which is provably not accurate. Why it would be more accurate in anything than for example MPC's model, so that it could correctly rate high quality codecs? There's just no proof it's better, only that it's worse with temporal accuracy.
Juha Laaksonheimo

MPC vs OGG

Reply #43 –
Quote
Originally posted by layer3maniac
In what alternate universe is an abx test blind? You KNOW what codec B was made with. That's NOT blind. If you're biased toward the psymodel codec B used, the results will reflect that. Some people are extremely sensitive to pre echo, some to dropouts, some to post echo, some to high or even low frequencies, some, like beatles, to nuances of overall depth or warmth, and some to stereo separation. ABX tests are skewed by ALL these biases, preferences, priorities and hearing difficulties.



Please read this document: http://mpeg.telecomitalialab.com/public/AAC_results.zip

This is a very good way to measure codec quality, and I think it is pretty much reliable.

MPC vs OGG

Reply #44 –
As an example of biases, preferences, priorities and hearing difficulties, Garf COULDN'T ABX this sample decoded from mpc. I COULD - 16 out of 16 times. However with ogg I couldn't. My results, like Garf's, are skewed by my individual biases, preferences, priorities and hearing difficulties.

MPC vs OGG

Reply #45 –
Quote
Originally posted by layer3maniac
In what alternate universe is an abx test blind? You KNOW what codec B was made with. That's NOT blind. If you're biased toward the psymodel codec B used, the results will reflect that.
What? ABX IS blind test. Sure, you know A and B, but you don't know X. That's what makes it blind. The idea is to find out if you can hear any difference. The only usable result from ABX is that you can say with high confidence that there's a difference. It doesn't matter if some people has not very good hearing and can't get high confidence ABX result as long as even ONE people can provide high confidence result that there's a difference. It's enough to prove there's an audible difference, and that's the purpose of ABX.
Juha Laaksonheimo

MPC vs OGG

Reply #46 –
Quote
What? ABX IS blind test. Sure, you know A and B, but you don't know X. That's what makes it blind. The idea is to find out if you can hear any difference. The only usable result from ABX is that you can say with high confidence that there's a difference. It doesn't matter if some people has not very good temporal hearing and can't get high confidence ABX result as long as even ONE people can provide high confidence result that there's a difference. It's enough to prove there's an audible difference, and that's the purpose of ABX.
It proves it to the individual listener doing the abx. To everyone else, abx proves NOTHING. Unlike an eaqual test, abx can NOT be replicated.

MPC vs OGG

Reply #47 –
Quote
Originally posted by layer3maniac
As an example of biases, preferences, priorities and hearing difficulties, Garf COULDN'T ABX this sample decoded from mpc. I COULD - 16 out of 16 times. However with ogg I couldn't. My results, like Garf's, are skewed by my individual biases, preferences, priorities and hearing difficulties.
I don't see anything biased in this. If you could truely ABX the mpc sample and Garf could ABX Vorbis sample, it means that these are both non-transparent.

When you start to speculate which one truely sounds better, then you are as much biased as anybody including EAQUAL's psymodel (which is biased towards certain response, but doesn't show fatigue problems)
Juha Laaksonheimo

MPC vs OGG

Reply #48 –
Quote
Originally posted by layer3maniac
It proves it to the individual listener doing the abx. To everyone else, abx proves NOTHING. Unlike an eaqual test, abx can NOT be replicated.
LOL , if you can ABX something with high confidence it DOES PROVE to everybody that the sample is not transparent.

But if you can't ABX the sample with high confidence that doesn't prove anything.
Juha Laaksonheimo

MPC vs OGG

Reply #49 –
Quote
LOL , if you can ABX something with high confidence it DOES PROVE to everybody that the sample is not transparent.

But if you can't ABX the sample with high confidence that doesn't prove anything.
As I said before, I DID ABX the mpc decode 16 out of 16 times. What does that prove? Can you duplicate it? Can Garf? That proves NOTHING. That's why I DID the eaqual test on THIS sample. EVERYONE can duplicate the eaqual test, which proves more than one individual's skewed abx test.