I would like to hear about the quality comparison results of the 4 formats mentioned in the subject.


edit: finally corrected the grammar...

I don't know of any comparisons like that with PsyTEL directly in the mix, but ff123 has performed a test with those formats and Liquid AAC instead of PsyTEL (which is regarded as performing slightly better than PsyTEL I believe).

You can find the analysis of the results here:

http://www.ff123.net/dogies/dogies_plots.html

Although a bit OT, from my personal experience (extensive blind listening tests), in relation to quality (and which ff123's test seems to support as well) the following codecs usually seem to come out in this order, from best to worst:

1. MPC
2. AAC
3. Vorbis
4. MP3

When RC3 hits, I suspect the gap between AAC and Vorbis will start to close. Right now AAC mostly has a big edge on Vorbis because Vorbis' pre-echo is very weak, possibly worse than a well tuned MP3 encoder even. MPC further edges out AAC in regards to issues like pre/post-echo and performance on difficult clips like impulses, etc. From a quality standpoint, MP3 should really only be used if compatibility is a very large factor to take into consideration, such as hardware support.

What I heard is that PsyTEL AAC (VBR) is better than Liquid Audio AAC with bitrates > 128kbps, I wonder whether this is true. By the way, is there any quality comparison results of full bitrate range (from very low bitrate to very high bitrate) of the audio formats?

Well, I feel MPC sounds lifeless comparing to PsyTEL AAC at high bitrates (Don't ask me for ABX tests, I never do) although MPC does handle very well on artifacts (I can't hear artifacts with -extreme) and has average lower bitrate than PsyTEL AAC (-archive) for every track encoded. I suggest to test the audio formats using really good speakers instead of earphones. Earphones can't show whether the music encoded is lifeless or not (unless you are going to blow your ear off with the earphone so close to your ears) although with earphones you are more likely to detect artifacts. And this is why I prefer PsyTEL AAC over MPC.

Ehh, ok. Now we have again a new very precise and pro definition, which really helps the developers...lifeless. Great, and no ABX or any proofs of course.

Please...

For newbies: Everytime you see someone say "I feel this sounds like that", the alarm bells should start ringing.
For testers these "feelings" are very familiar...

Just earlier today I was talking about these very types of issues..

It seems there are some common underlying themes to various claims that I keep seeing pop up that people are not willing to back up. You see them most commonly in a very vague description of the sound. Here are some usual examples:

1. <codec> lacks bass.
2. <codec> lacks highs.
3. <codec> sounds colder.
4. <codec> sounds warmer.
5. <codec> sounds dull.
6. <codec> sounds shrill.
7. <codec> has some weird distortion that I can't explain.

etc.

Now this one falls right in to that category -- <codec> sounds lifeless.

Honestly.. if people are so sure that they can hear something, enough to where they would avoid using it, then why wouldn't they be able to prove it? Or why wouldn't they be willing to try?

At any rate, the type of descriptions I mentioned above, as JohnV says, are the usual signs that something is amiss. Usually psychoacoustic audio compression does not behave in a manner similar to what you would expect from conventional audio and the differences you might hear between systems/media/etc.

In psychoacoustic audio, you don't really get "warm" or "cold". You get "artifacted" or "not artifacted". You for the most part shouldn't get things like a "lack of bass", though you might get distorted bass. With the exception of lowpasses and highpasses used by various encoders, usually you aren't going to be actually missing something from the sound, you will more than likely have things "added to it", perceptually, in the form of artifacts. You usually don't get subtle changes in tonality or the general "feel" of the audio, instead you get distortion of the details. Pre-echo, Post-echo, dropouts, sometimes a distortion of the stereo field, etc.



Testing audio codecs with speakers is fine, testing with speakers over headphones however makes little sense to me, unless you want to set yourself up to purposely not hear certain aspects of the sound. It just sounds like a way to set things up to hear what you want to hear, and not hear what you don't want to hear.

However, I completely disagree with the whole "earphones can't show you whether the music encoded is lifeless or not" part. First of all, define "lifeless"? Since we are talking about audio quality, I'll assume that it is related to "not sounding good". In that case, listening with headphones certainly can tell you whether or not something sounds lifeless. To me lifeless is something that sounds artificial. Something that I can actually quantify and relate to artifacting. A good example of this would be WMA. WMA has "lifeless" highs, they sound artificial and synthesized, mp3pro is the same way -- these are things that I can abx and these are both things I certainly can hear on headphones.

Anyway.. no offense to you or your opinion, but there have been so many people lately saying that MPC sounds bad or has some sort of mysterious flaw (and it's always a different story I might add) now that the format starting to gain in popularity more, yet not a single one has provided a shred of evidence of this yet, at all. Period. It's all hearsay and speculation, and now the trend actually seems to be that people want to make these statements but are unwilling to even attempt to back them up. That means to me that they probably realize they are wrong, or they are so unsure of their statements that they don't want to lose face if they can't deliver the results. Again.. if either of these things are the case, then why bother in the first place?

Just my 2 cents.. and btw, none of this is exclusive to MPC either, statements of these sort made about any codec should really be backed up with some sort of evidence.. that is at least if there is to be progress made.

Dibrom,
What if the problem isn't with the encoder or decoder but the plugin? ABXing wavs would do nothing to expose a problem like this.

So decode the file to .wav via the plugin instead of the command line decoder. Problem solved.

Liquid Audio can't be decoded with the plugin. Exactly how did ffc do that anyway? Burn it to a cd and then extract the wav? Hardly a fair method, wouldn't you say? Unless he did that to all the formats.

He used total recorder which just intercepts the audio before it hits the wave device. It provides a perfect digital copy of the audio being played back, in the same manner that something like the diskwriter in winamp would.

For comparison tests, I use a winamp plugin, in_lqt.dll version 1.055, to decode Liquid Audio lqt files (as documented on my pages). I believe this is the decoder usually used by the people in alt.binaries.sounds.aac.

The only codec I used Total Recorder on was WMA8 because there is no "write to disk" function available in WMP 7.1, and I wanted to quell any doubts about this or that decoder (i.e., in_wma.dll in Winamp 2.6) not being the "correct" one to use -- the Winamp and WMP wma decoders differ by a couple of bits.

I was also forced to burn the wma file to CD to encode, because that was the only way Windows Media Player 7.1 allowed me to make a wma file. I could have used wm8eutil.exe, but again there are slight differences in the two encoders (both WMA8!) and I wanted to encode in the way that most other people probably do.

ff123

I'm not going to send any clips or anything of that nature, but I'm going to describe my experiences with mpp/mpc. I started using the format many years ago because I liked the fact that there were no pre/post echo problems. As far as I was concerned, and still am, there are no artifacting problems with the format which is quite an impressive thing.

One day I decided to compare a few formats by burning a cd with the original wav as well as the decompressed formats in a few musical genres. I was quite surprised to find what I thought might be tonality problems with mpc. At first I thought it was just a placebo effect, I had heard others complain of mpc sounding flat in the high end and thought I might have been predeposed to 'hear' flatness in the high end. That was utter crap as far as I was concerned. If anything I would have considered mpc to sound slightly brighter in the high end frequencies than the original.

I am not going to pull out samples for you to double blind test. I have a terribly slow dial-up connection, as well as no burning desire to do so. However, the type of music I was listening to that seemed the most changed was r'n'b/pop vocal music and it sounded as though the timbre of the notes had been stripped from the original. I never found this while listening through computer speakers or a good pair of head sets, but I thought it was prominent enough on a full stereo system to mention it.

Typically I encoded in x-treme

Cool I just downloaded the 1.055 liquid audio plugin. Thanks for the info!

JohnV, I know you are 1 of the expert listeners and well trained to detect artifacts. But I think you side too much on MPC and feel that other formats are just craps (psychological problem). When you tend to side something, you would say others are bad. Okay, I clarify that I am not trying to insult any people here.



In fact I am not a newbie. Very sorry to tell you so if you are trying to insult me as a newbie. I had tried many audio formats (MP3Pro, LAME MP3, Fhg MP3, RM8, WMA8, MPC, AAC, etc) by myself. I had visited other listening results web sites as well, e.g. ff123's (I would like to know about what other people think about the best 128kbps audio format).

Listening is very subjective and I am just voicing my opinion that MPC in my ears not as good as PsyTEL AAC in terms of life of the music. I do agree that MPC gives no artifacts (-xtreme for my ears, have not able to find any music to kill it yet. PsyTEL AAC -archive has no artifacts for me as well), handle perfect on transients, perfect psychoaccoustic model and it is great format (I do agree with this, in fact every invention of audio format is great) but it sounds lifeless compare at the almost the same bitrate (average 10-50kbps difference), MPC -xtreme vs PsyTEL AAC -archive (again to my ears and perhaps to NitTheBlak's too, for his case is MP3. This is very personal).

I never say everybody should use PsyTEL AAC like those people who are still going around to convince people to use their old format VQF. I just said that I feel PsyTEL AAC is good in terms of life of music. And then you people are trying to shoot me on this Eh I want opinions only, you so fast conclude I am a newbie??? It really sounds like brain washing. Even if the person is a newbie, you should not say they are all the same, in fact they may detect other things that you might not able to detect.

Just to point out that artifacts and life of music both are different, you could not say lifeless means there is artifact. To be said clearly, lifeless means you always feel that there is lack of something in the music perhaps the psychoaccoustic model removes too many details e.g. masked too much, etc (e.g. NitTheBlak said it's on the mid frequencies, for my case I use my feelings to feel it). Artifacts means you have extra something (ringing, flanging, etc) due to lossy compression comparing to the orginal. When you tend to pay more attention on detecting on whether there are artifacts, you would have less attention on whether the music is lack of something.

And yeah, by the way, is there any more version updates for PsyTEL AAC?

Yes, artifacts is like there something added on the music due to the lack of bitrates to represent the waves. I do agree with you on this. I had mentioned in previous post as well.



Good speakers would let you hear not only the increased of loudness, you could even get real ambience as well, and you wouldn't miss a single tone also as you'd mentioned. If headphones are really that good, then why you still go for live concerts? And with the real ambience, loudness, etc (I am lousy on describing and my English is not that good) there you could really feel the life of the music (YES, LIFE OF MUSIC!!!), unless you haven't been to live concerts before. Try to compare the audio formats with a pair of decent stereo speakers, at least not earphones. I do agree that headphones are pretty good to detect artifacts.



Okay, I think we have a misunderstanding here. Your definition of "lifeless" and mine is different. Mine is explained previously.

I do agree that WMA sounds real metallic (artificial and synthesized as you mentioned) and I wonder M$ consider 64kbps WMA 8 is CD quality (craps). I feel that MP3Pro is better than WMA 8 for 64kbps.



I am not here to say MPC is not good. In fact every audio format is great (they all try to achieve a good size/quality ratio for digital audio). I think I might has chosen to wrong timing as well, as other people are trying to bash MPC.



Evidence is yourself. You are the listener who decides what sounds better to your ears.

Whoa... hold on there. You have some [b]HUGE misconceptions man. I don't think I've ever seen JohnV say AAC or PsyTEL was crap, quite the contrary I'm almost positive he recognizes PsyTEL (as I do) as a very high quality encoder. For that matter he also liked Garf's tuned Vorbis encoder quite a bit. So please, try to get your story right here. JohnV's interest in MPC isn't psychological, its because MPC performs extremely well. JohnV actually has participated in helping tune PsyTEL (and many other codecs) to my knowledge, so before you go off saying that you think he is biased and has some psychological problems, you really need to check your facts. You are so far off it isn't even funny.



First of all, I don't buy this. Sure, audio quality is subjective, but psychoacoustic audio quality isn't nearly as subjective as some people would like to think. When I see arguments like "I prefer MP3 artifacts over MPC artifacts" then I have to ask... where are these MPC artifacts? Show me. Provide a clip for me, and prove that you can hear a difference. You don't like calling that difference an "artifact"? Fine, but you still claim you hear a problem.. so prove it.



Fine.. say I was to give you the benefit of the doubt -- to you it sounds lifeless. So you should have no problem hearing this "lifeless" quality in a blind listening test right? Because if you can't even on a clip of your own choosing, then you aren't hearing a difference.



And we never said that PsyTEL was bad. I'll be the first to admit, it's an awesome encoder. If there was no MPC, I'd use PsyTEL for my highest quality encodings.



Another misconception. We are not trying to shoot you down on your statement that PsyTEL sounds good. I agree 100% with you there. What I don't agree with is your statement that MPC is lifeless. I'm inclined to believe people's claims normally, but when someone is so unwilling to provide a test clip or any listening test results, that means to me that what they say should be considered highly suspect.

You say that some people are able to detect things that others are not. I agree with you 100% there also. No doubt about that at all. And again, that brings me back to my point, if you can detect this issue that you claim you can, then why are you unwilling to perform a blind listening test?



You use your "feelings" to feel it? That sounds to me like you are convincing yourself you are hearing something that you are not. Somewhat ironic given the fact that you would portray JohnV who is absolutely willing to always back up his statements with hard data, as the one who is affected by a psychological issue in relation to perception.



Not true. Due to the nature of psychoacoustic audio compression, when you are listening for artifacts, you are listening for any difference, whatever that may happen to be. You don't separate these two issues out as you would seem to imply.

So in closing, sure it's possible you are hearing something. Anything is possible. Certainly if there is a problem it should be uncovered and publicized so that it can be addressed. However, when you refuse to participate in a listening test, provide any data, and instead just make wild claims, describing this "problem" so vaguely, it means absolutely nothing.

I found out that PsyTEL AAC has this problem too, for modes below -extreme (or -normal?). For test clips, please use real music (some minutes long), it is not difficult to find out the problem.



Same here, I usually encode MPC using -xtreme.

First of all, you don't need sheer volume to hear differences in audio codecs. In fact, the masking effect is increased at higher volumes, so unless you are listening to an extremely quiet sample, it's even less likely you'd hear a difference at higher volumes. So then you have 2 things, you aren't going to hear the differences in detail as well on speakers, and if you are using speakers to get increased volume, you aren't going to hear further subtle differences which would normally be audible at lower volumes.



I go to concerts to see bands perform. Most of the time the sound quality of their performances suck compared to some well produced studio content played back on a decent system.

I don't go to concerts because of the audio quality, I go to concerts for the experience of being there, in that it is a social experience.. the two are totally unrelated and have nothing to do with the "feeling" of the sound.



Anything which you say you can hear or "feel" you should be able to do so reliably, so there should be no problem for you to detect this "life of the music" in a blind listening test. If you are so sure you can hear all of this, then why are we arguing? Why the refusal to abx?



I personally don't feel they are all great, though the idea behind psychoacoustics and the fact that it works is pretty amazing I'll admit.



Exactly, but this is not a reason not to perform blind listening tests. Because you should be able to hear the difference reliably yourself at the very least.

If you want to "define" your own truth to yourself and convince yourself that everyone else is wrong (ignoring the fact that you may not even be able to reliably tell a difference yourself because you have not tested), that's fine. Everyone is entitled to do that, but it doesn't help anyone in a community working towards achieving higher quality audio compression.. if anything it only spreads misinformation and confusion.

Since it is not difficult for you to find a piece of music which causes this problem apparently (now it sounds as in both PsyTEL and MPC) then please copy the relevant portions of a track, encode with LPAC, and upload it somewhere so we can all hear these alleged problems.

If you refuse to do that, then tell us the exact track on the exact album along with some timecodes so that someone who has this music can help us out instead.

Thanks for the information. I would like to apologize for my misconceptions.



By the way, I did not say "I prefer MP3 artifacts over MPC artifacts". Don't generalize me with others, please. I do agree that I can't find any artifacts for MPC yet (Why yet? God knows 1 day some people could manage to find other killer music).



Okay, I forget to mention that my listening tests that I did by myself is like using shuffle mode in Winamp and play all the encoded audio formats with the original in the same playlist without me looking on which format it is playing. I wonder whether you could call this as blind listening test.



I never say you guys said PsyTEL is bad. Maybe you have misread my words.



The definition of "lifeless" of yours and mine are different. I think we have misunderstanding here. My test clips are just real musics. When you blast real musics on your decent stereo speakers, you would notice them. I do agree that the test samples for tuning psychoaccoustic models are real musics too but the thing is real music last for minutes and samples last for seconds. The length could tell a difference because our brain limitation.



Feeling to feel it does not mean I am trying to convince myself. I think you have misconceptions about this. Do you try to convince yourself if you feel that you like a girl??? No right?



The "adding" means those things that the original wave doesn't have e.g. ringing, flanging, etc (artifacts).



To clarify, I am not making wild claims. Maybe we have lots of communication barriers for this (my English is not good).

Not true. The audio reproduction of speakers and earphones are different. This is another factor of audio quality that we shouldn't miss out (other factors are like DAC converter of sound card, etc).



It does have, argh. Anyway, I do not know how to describe to you (my limitation of using English to describe). I am not trying to convince you to follow me. I just want to hold my stand.



I had said how I do my listening tests in the previous posts.



Well, I agree that lossless compression at anytime is better than lossy compression because you would not get any bits lost during the encoding process. You would not need to bother the flaws of the psychoaccoustic model as well.



I am not here to define my own truth and I am not to convince myself that everyone is wrong. I am here to discuss, to exchange and share knowledges. To clarify, I am not here to spread misinformation and confusion as well.

The problem is that not everyone in the world is with a broadband connection.

For the musics, I listen to Chinese musics.

The comment was made to illustrate my point that some people make up excuses why they prefer one format over another, aside from all logic or established data that exists.

You say that you prefer PsyTEL because MPC sounds lifeless. Show me a sample where it sounds lifeless, give me some evidence.

QUOTE

Okay, I forget to mention that my listening tests that I did by myself is like using shuffle mode in Winamp and play all the encoded audio formats with the original in the same playlist without me looking on which format it is playing. I wonder whether you could call this as blind listening test.

First of all that is not blind testing, and second it is rather convenient to say after the fact. It is very possible that you have done some sort of conclusive testing, however until you share your results in the form of a test clip or some provide some sort of reproducible evidence, or even a desire to do so, then your claims hold no water.

QUOTE

I never say you guys said PsyTEL is bad. Maybe you have misread my words.

I didn't misread your words, though it appears you have missed the point that I was trying to make. You came across as if you thought we were attacking you for stating that you preferred PsyTEL. I explained that this was not the case, and instead our responses were related to your unsubstantiated claim that "MPC sounds lifeless".

QUOTE

The definition of \"lifeless\" of yours and mine are different. I think we have misunderstanding here.

There is no misunderstanding. The problem is that what you call "lifeless" you cannot (apparently) quantify, nor are you willing to provide a single bit of evidence that shows this behavior.

QUOTE

My test clips are just real musics.

So are all of mine. And for that matter castanets, fatboy, spahm, all the killer clips are "real music". They may not be indicative of the majority of music, but rest assured that they are from [b]real music.



Again I ask, what music? What track off of what album, and at what time?

QUOTE

I do agree that the test samples for tuning psychoaccoustic models are real musics too but the thing is real music last for minutes and samples last for seconds. The length could tell a difference because our brain limitation.

Would it be better then if I provided a sample harder to encode than fatboy, but that lasted nearly 10 minutes?

QUOTE

Feeling to feel it does not mean I am trying to convince myself.

If you only say you can "feel" something, but you can not quantify it, and you cannot reliably prove that you can detect it, then it cannot be considered that you are actually perceiving any difference. That's just the way things work from an objective approach.

QUOTE

I think you have misconceptions about this. Do you try to convince yourself if you feel that you like a girl??? No right?

First of all, the whole point of public listening tests is to gather objective data. Something such as emotions are not objective at all. Furthermore, they are completely unrelated to the discussion at hand.

You can argue that you prefer the sound of PsyTEL because of its possible quirks and modifications it makes to the sound, but you run into problems when you try to imply that that somehow relates to encoding accuracy. It does not.

You can say you prefer the sound of vinyl for example, but that does [b]NOT mean that CD Audio is not accurate, it simply means you prefer the distortion that vinyl is subject to.

However.. I'd still like you to prove that you can actually hear the difference enough to reliably pick one as "feeling" better to you than the other. As it is, right now I do not believe you because you have not given any sort of data to validate this claim and it just looks like you are making excuses on why you shouldn't need to.



They are wild claims until they have been somewhat substantiated.

Did you miss this:



So give us some details already.

Not true? Would you like me to provide you with clips that exemplify the fact that the masking effect is greater at loud volumes? I'd be more than happy to.

QUOTE

It does have, argh. Anyway, I do not know how to describe to you (my limitation of using English to describe). I am not trying to convince you to follow me. I just want to hold my stand.

If you can't describe it, you can't quantify it, and you can't prove it... has it ever crossed your mind that it doesn't exist? Heh..

From an objective point of view, one simply cannot just "take things for granted" like that.

Anyway, I stick with my comments that concerts and psychoacoustic audio compression are two entirely different things. Taking the discussion off track doesn't change anything.

QUOTE

I had said how I do my listening tests in the previous posts.

And they are not blind, objective, or reproducible by anyone else so far.

QUOTE

I am not here to define my own truth and I am not to convince myself that everyone is wrong. I am here to discuss, to exchange and share knowledges.

Ok, then I'll ask again. Where is a clip that when encoded shows the behavior you described? If you cannot upload a clip then give use [b]exact details on where we can find this clip on a real album elsewhere.

If you are here to share and discuss as you imply, then why not share and discuss this with us?



Refer to the previous statement.

Just thought you'd like to hear my ABX results so far.
I started with a file I encoded with Liquifier Pro 5 at 128 & 192. I decoded them with Liquidy Split 1.06b. It was easy to tell them apart from the original wav, 11 for 11. Then I encoded the same file with mpc at radio which averaged at 145 and extreme which averaged 185, honestly I couldn't tell either of them from the original. The mpc even at 145 was FAR superior to the lqt at 192. I'm going to do some more testing with other encoders, I'm encoding the same file with PsyTel 1.2 right now, but I have to say that I found mpc even with just the radio preset to be VERY impressive.

layer3maniac:

Interesting results.. and could be possible depending on the clip. I think that a few people are going to want to have a reference of what you are encoding to try and reproduce test results though (standard procedure). Can you maybe provide a link?

Well, there must be a problem with the in_lqt plugin I'm using because the decoded PsyTEL file even at 114 sounds excellent compared to the decoded lqt at 192. I'll try earlier versions but I really wish ff123 had used PsyTEL for aac in his test now... Ivan sure knows how to make a 114 file sound good. This isn't fatboy.wav but I don't listen to fatboy slim that much anyway. Right about now... da funk soul brover

Sure thing Dibrom, but first I want to encode some more challenging files. Not really encoder busting files though, just some "busier" music. When I get some reliable results you will be the first to know. I'm anxious to test out some files encoded with Garth's tuned ogg encoder too.

Hmm... I have to join this interesting discussion

First, my comments on MPC - MPC is very high quality audio encoder, probably best in its range. Andree has done quite remarkable work on psychoacoustic model optimization and applying psychoacoustic model suitable for transform coders (such as AAC) to one subband-only coder, like MPC. I have learned many things from his work and his comments/advices of my psychoacoustic model implementation.

When I started my work on AAC psychoacoustic optimization, MPC was near in production state, so I had hard time to "catch-up" with the MPC quality. Final result is the AACEnc -archive sounds very similar to MPC -extreme.

MPC has less "problem" cases than PsyTEL AAC on -extreme and -normal presets, but most people would agree than in MOST cases they both are high quality. In some cases, AACEnc might generate small drop-outs (on -extreme and -normal values) that are not recognized by most listeners. This is due to some internal design trade-off in psytel AAC which I can't modify now, and AAC filterbank design, too (this is not fixable). However, -archive value is very good, as good as MPC.

Now, MPC certainly does not sound dull to me, at least I haven't found any clip that does sound "dull" or something like that to me.

Pros of MP+:
- Transparent quality at -extreme setting for 99.9% samples
- Low Decoder Complexity, lower than MP3 and AAC
- Excellent psychoacoustic tuning

Cons of MP+
- Not very flexible at lower bitrates
- No CBR support (can't be used for straming with small delay)
- Not a ISO standard

Pros of AAC:
- Transparent quality at -archive (psytel), very near at -extreme
- Good CBR support with high audio quality (better than MP3)
- ISO Standard
- Many sampling rates (8000-96000 Hz), up to 256 kbits/s per channel, up to 48 channels

Cons of AAC:
- Problem cases that trip out all transform codecs
- Dropouts sometimes possible (psytel aac)
- Very tight licensing schedule (ISO AAC is not for end-users)
- Increased complexity

No, I don't consider other formats craps. Well, Dibrom already have said pretty much everything, so I don't need to write long message...

No, I'm not trying to insult you and I didn't say you are a newbie. I was warning newbies not to believe everything they see is said. The sad thing is that the psychological factor is very great here, meaning if some people would do now non-blind testing, they would probably try to hear the "lifeless" sound of MPC, and because of the psychological factor they really think they hear something like this. Only proper blind testing reveals this may not be true.

Well, ABX testing is totally objective. After you have reliably verified you can hear a difference, then you can start to speculate what is it you hear. It's not always practical to do full ABX testings when tweaking codecs, then for example blind AB-test is often used, but the key point also here is repetition. You gotta get the same result many times.
You can download ABX comparator from here:
http://www.pcabx.com/program/ABX173_setup.exe

With blind testing the point is just to verify you can hear a difference. It doesn't matter what you call it; distortion,artifacts or what ever. The point is to verify you can hear a difference. The word "difference" here covers everything: single artifacts, constant distortion, excessive noisyness, even the effect you would call "lifeless", everything.

What I would suggest is that you would do ABX testings between both original vs MPC/Psytel and Psytel vs MPC. Also use fair comparable settings (bitrate) for both.

You should be able to ABX this difference then.
I hope you read also the experiences of Niktheblak:
http://www.hydrogenaudio.org/forums/showth...s=&threadid=325

Lastly, I'm not saying it's impossible you are hearing this "lifeless" sound. But as long as there are no proofs (your ABX results, test clip), these opinions can't be taken very seriously and the developers have no chance to try to identify and correct the situation.

I see some similarities in our claims and I have to respond just out of the curiosity value of this issue. As a guitarist I pay very much attention to electric guitar sound and that (and that only) is where I thought that dropping might have happened. In a feeling-domain it could be described as a lack of PAIN with PAIN meaning that when listening to a distorted guitar at reasonably high volumes it sounds somewhat PAINful (the ear reacts to high-intensity mid-frequency sounds with a sensation of pain). My claim aroused from the experience that MPC version of the distorted guitar sounded less PAINful.

In frequency domain it might be seen as attenuation of some mid-range frequencies the ear is most sensitive to (or which trigger the painful sensation). Theoretically that could happen since the frequency spectrum of a distorted guitar is very wide and the encoder would have to dispense the limited amount of bits equally (according to the ATH, that is!) among different frequencies. Especially if plenty of other high-intensity mid-range sounds are present at the sample.

Theoretically high-intensity wide-frequency range sound consisting mostly of noise is rather difficult to encode and I also remember that most codecs indeed have problems with distorted material. That's why metal yields much higher bitrate than "easier" material with certain codecs.

But with MPC -standard everything is ~175 kb/s. Dibrom's standard preset encodes my metal with ~260 kb/s, GTuned2's 160 mode (-b 999) gives ~210kb/s while oggenc -b 256 gives ~300kb/s. But MPC -standard stubbornly encodes everything I feed it with at ~175kb/s. So either some high-order compromises are made or MPC's psymodel is truly superior. My ABX tests indicated the latter.

You have to remember that all I have said here is strictly hypothetical and it should not be viewed so much as a claim as rattlings of a frustrated sceptic.

Althought I am pretty sure of MPC's superiority I will remain a bit cautious with this matter. The instant I spot a decent test clip I will do a series of ABX's and perhaps present some nice frequency response curves hopefully showing that some dropping actually happens. Until then I would recommend happy MPC'ing

I'd certainly agree with all of this. For the vast majority of situations, PsyTEL is very high quality indeed, transparent even.

One of the issues that I'm faced with (and I realize this doesn't apply to probably even 1/8th of the people) is that at least half of the "standard" music that I encode and listen to happens to be of the type that cause these very problems with transform codecs. However, I believe PsyTEL AAC handles them better than any other transform coder I've seen yet, which says a lot really. I also realize that (as Ivan says) many of these differences approach a level that most listeners will not recognize.

However, given that MPC does edge out AAC on these pathological cases, and that it is free, and it is very fast, it just makes sense for me to use that format. I'd really love to use AAC more (since it seems to be gaining hardware support), but the fact that it is difficult to use the format as an end user makes for a bad situation unfortunately..



I believe that encoder speed should be added to this list. AFAIK MPC is also the fastest encoder out there for the level of quality that it provides.

I pretty much agree with everything else in your post also.

If you want my opinion on this matter, I suspect "similarities" between all these types of cases are nothing more than people reading there may be a problem, then somehow convincing themselves there is a problem and that they are actually hearing it. This effect is not really so uncommon..

I'm not saying this is what happened with you, or with VeryBlur or anyone in particular.. but the fact that claims keep popping up with no results and everyone is so unwilling to do tests or try to prove any of it, (except you, which was a nice change) I think supports this possibility.

QUOTE

Theoretically high-intensity wide-frequency range sound consisting mostly of noise is rather difficult to encode and I also remember that most codecs indeed have problems with distorted material.  That's why metal yields much higher bitrate than \"easier\" material with certain codecs.

But with MPC -standard everything is ~175 kb/s. Dibrom's standard preset encodes my metal with ~260 kb/s, GTuned2's 160 mode (-b 999) gives ~210kb/s while oggenc -b 256 gives ~300kb/s. But MPC -standard stubbornly encodes everything I feed it with at  ~175kb/s. So either some high-order compromises are made or MPC's psymodel is truly superior. My ABX tests indicated the latter.

Well as far as MP3 goes, the only reason you are seeing bitrates that high is because of a [b]flaw in the format. It cannot encode content above 16khz efficiently at all. If you want proof of this effect, encode the file again with the -Y switch which disables noise shaping in this region, basically keeping vbr from trying to increase bitrate to compensate for the fact that it cannot encode this content accurately otherwise, because of a lack of a scalefactor for the last band.

If you encode with -Y you will see the bitrate drop significantly.

As for Garf's mode coming out at 210kbps, you can't compare that directly to MPC for a number of reasons. First of all MPC exploits more psychoacoustic effects than vorbis does (temporal masking for example) and I also believe that the lossless stage in MPC might be superior to that of Vorbis (I'm sure Garf will correctly me here if I'm wrong), so that saves even more bits. All of that and MPC has also had the benefit of being tuned "to the wire" much more than Vorbis has.. so you gain even more efficiency.

And as for -b256, you can't compare that to MPC standard because you are forcing a higher bitrate in Vorbis on purpose. So to have any sort of valid comparison you need to force a higher mode in MPC as well.. try -insane for example, if you just want to see higher bitrates.

Lossless and saving bits are a bit contradictory arent they

If you're talking about the transform stage, Vorbis has the edge there because the MDCT is superior in terms of efficiency to subbanding (hence Vorbis will beat MPC at lot bitrates). The problem of the MDCT is that it's more susceptible to preecho, which then may cause a need for more accurate encoding in the lossy steps, causing MPC to win out in those situations.

Overall, I find it a bit hard and irrelevant to say what codec is better based on the internal steps it does. MP3 does both subbanding _and_ MDCT so it should easily beat Vorbis and MPC, right

Bitrate fluctuations on passages are a bad way to judge a codec on. Sometimes an encoder will use a very low bitrate on a hard passage and get away with it without problems. I often see that passages that I think are hard to encode really are not. It's hard to predict bitrates. You certainly should not assume that because a codec uses a low bitrate on a passage and others do not that there is a higher chances that it will artifact there. Quite the contrary: it indicates it can encode the passage well when the others cannot.

As for the rest of the thread, we've had a zillion of these already and they unvariably end up the same. I don't have the patience like Dibrom to explain everything for the umpteenth time, so I will be very short and blunt:

Either you provide a sample or point out the CD and track, or you shut the fuck up.

--
GCP

I believe Dibrom means MPC's lossless phase: huffman and differential coding and maybe some other lossless methods.

My choice of Liquid Audio for my 128 kbit/s comparisons was based on listening tests I had performed for myself using Psytel 1.0:

See:

http://ff123.net/peaceful.html
http://ff123.net/duel.html

I believe psytel sounded better on duel.wav, but was quite bad on peaceful.wav.

Also, back then, bAdDuDeX complained of ringing in psytel at 128 (which he cured for himself by lowering the ATH, I believe). Liquid Audio 128VBR was actually his choice of best low-bitrate encoder, even though that setting is not really low bitrate (can go up to 160 kbit/s on some music).

Psytel 1.0 is an older version, though, and Ivan has improved CBR 128 since then. Ivan's tests of castanets.wav (see results of 128 test here: http://ff123.net/export/aac128log.txt) appears to show that Psytel is very close to Liquifier quality. I participated in that test, and my personal results was that FhG (Liquid Audio) was clear and away the best on castanets.wav at 128.

ff123

Not quite - MP3 is using subbanding and MDCT because MPEG commitee needed to keep some degree of filterbank compatibility with MP2, don't ask me why - I don't know And, at the time MP3 was standardized pure MDCT codecs were not yet superior to subband ones at low bitrates, and important companies were "pushing' subband design.

In that case, I would expect Oggs VQ to do better than MPC's quantization+huffman.

But I have no idea, really. (looks like apples and oranges to me)

--
GCP

Of course not. I was being ironic. Maybe you missed the ?

My point was that the actual transform method does not provide any guarantee whatsoever for audio quality.

--
GCP

Yes, this is what I was talking about.



Could be. I was just trying to find ways to explain why MPC would still outperform Vorbis at a lower bitrate, even with the GT modes being used. I know that the lossless stage in MPC is really very efficient (and I assume Vorbis' is as well)and it might be a way that MPC could saving a few more bits. Of course most of the other stuff mentioned is probably more likely, temporal masking, more fine grained tuning, etc.

I'll just wade into this little debate. I've been a recording engineer for 25 years and an audiophile for a bit longer than that with an extremely high end system. Currently testing Musepack, AAC and FAAC. At this point in time they are truly the only ones that would be in the running for discerning listeners. LAME unfortunately destroys the music to the point where it could not be used for any critical listening. Currently comparing using an Al Stewart track that I was present for the recording of so I know what it should sound like. If anyone's interested I'll let you know my findings once I'm done. I certainly wouldn't, at this early point, consider any of the formats suitable for archiving.

Ehm, FAAC for discerning listener??? Well, anyway, what kind of settings and bitrates you use for testing?
Many of us are interested in your findings, so please share when you are done.

Originally posted by Beatles:


I am very interested in what you think of lame's quality if you use
the following commandline:

--alt-preset normal

and this version of the lame encoder, namely lame_dm_rev6.zip:

http://static.hydrogenaudio.org/extra/lame_dm_rev6.zip

What part of the music do you feel is destroyed? Content above 16 Kilohertz? Transients?



Not even MPC musepak/mpegplus at the -insane setting?

No no FAAC is clearly not an alternative, that was QUICKLY obvious. I am only interested John in absolutely the highest quality settings. Smaller bitrates are not any concern to me although it may be fun to tweak some to see just how low acceptable results can found. I've been playing with many different settings and really don't want to report what I think until I've fully tested everything. I see too many people shooting from the hip without being able to back up their findings. Two little tidbits I find interesting:
Listen to the rhythmic propulsion of a track known in Linn talk as the "foot tapping" ability of each encoder. I do find differences between AAC and Musepack with one clearly better but this is just preliminary.
Also in decay and overtones which allows one to hear soundstage cues and to hear into the music and discern the back and side walls I can hear a difference.
MP3 is clearly not in the same category as Musepack and AAC.
As I'm new here is the IRC channel a good place to discuss things?
Thanks.

I am using REV6 but don't consider MP3 as a suitable medium for anything other than sharing or casual background listening but that's just MY opinion.


What part of the music do you feel is destroyed? Content above 16 Kilohertz? Transients?

Mainly overtones, decay, sense of space and propulsion. I suppose obviously palpability could be added to that list.



Not even MPC musepak/mpegplus at the -insane setting?


No not for archiving. But archiving to me means would I save my masters in a certain format and no I personally wouldn't even consider it. But both Musepack and AAC are eminently listenable formats. I found Muse to be very good but I was was using an absolutely ludicrous commandline that gave an extremely large filesize and bitrate but sounded quite good in most areas. Overtones and soundstage flattening excepted.

Dear Beatles,

Thanks for your comments.
However, for us non-engineers can you explain (briefly) in layman's terms what you mean by some of the jargon?

For example,
overtones, decay, sense of space and propulsion, palpability

Here are some guesses (probably incorrect):

1. overtones: as guitarists know when you hit an e note there is also an overtone (b note) that is present though in much much less volume...

2. decay: rate at which notes fade out?

3. space: stereo image and its characteristics, perhps an mp3 can take a cathedral sound stereo image and dampen it so it sounds more like a medium room?

4. propulsion?? no idea...

5. palpability?? no idea

Interesting. Btw what kind of commandlines you used for MPC, I guess you also disabled mid/side-stereo? I could give you some hints for very high bitrate/quality commandline.
I don't think there's much use to go over 600-700kbps because then lossless coding becomes a viable choise.

Yeah, I wouldn't save masters in lossy format either...

Thanks for your excellent replies, Beatles!

Its nice to talk to an audio engineer...

Could you recommend some songs that really show off propulsion and palpability so i can do some tests?

Also have you ever tried --alt-preset insane with lame_dm_rev6?

I know there are problems with the mp3 format in general, but I'm curious to seek if you notice that --alt-preset insane is considerably better than -alt-preset normal... if you find that insane is not much better then it seems to me that normal is the way to go if someone is going to use mp3 at all....

Thanks,
AND HAPPY THANKSGIVING EVERYONE...