I have read at r3mix.net and at some other places that you can not hear any difference between CD sound and MP3 @ >=256 kbit/s. Great i thought. Let's convert my CD collection to high quality MP3 then.
If 320 kbit/s is needed i don't care as long as the sound sounds the same.

So now the problem.
I did a listening test of course.
I used Exact Audio Copy to rip my Haddaway "what is love" single.
Tested on my favorite song What Is Love (Club Mix).
I listened carefully to the first seconds of the wav file and the mp3 file in winamp many times.
I found a little difference in the lower frequencies. Every time the same clearly hearable difference. And you must be death if you can't hear this difference. The high frequency sounds the same but the lower frequencies that are played at the same time are less "rich" and loud in the mp3 version.
I tried the --r3mix -b112 preset "flags" as well as CBR 320kbit/s (highest possible right). They sounded alike but none had the same sound as the wav file. Is there a bug in the winamp player or in lame or what? Or is this talking about as good as CD just crap.

Appreciate any answer to this.
I use good earphones with excellent low frequency reproduction.
It's a pair of Pro.2 model PH-333. (Sorry i should not do commercial here but if somebody else want to try, it might be important).
Frequency range 3-28000 Hz.

I have used Razor Lame 1.1.5 with Lame 3.90.3 (and also tested with 3.93.1).
No sound difference between these either.

The sound card is a normal built in one. Nothing fancy at all.


------
I think i found the answer now. Also I'm in the wrong topic sorry.
I use the equalizer to dampen bass because it's too much in my earphones.
If i deselect the equalizer i don't hear any difference in the sound.
But now it's harder to tell because theres so much bass.

i have been noticing some bad things with LAME lately. i can hear artifacts in every song encoded with -alt--preset standard. so after deleteing my collection i encoded with --alt-preset cbr 256 and couldn't tell the difference between the wav. and the mp3. i too have the soundcard that came with the computer, maybe this has something to do with it. oh yeah i'm using 3.90.2 which is suppose to be the best version of lame.

there is one song that stands out as having the worst quality with -alt--preset standard, Exit Music (For A Film) - Radiohead. At 25secs or so when he starts singing his voice sounds very gargley. the wav. file sounds nothing like this.

this probably doesn't answer your question but i figured i would say it anyway.

The --r3mix preset is really outdated. You should try --alt preset standard and see if you can still hear the difference.

1- You are correct it's impossible
2- Use a lossless format
3- Or use MPC or OGG they sound better than Mp3 ( To Me )

i would be willing to try ogg if i knew the best settings or what programs to use. i have tried it once but i had no idea what i was doing and the file sounded like it had no bass.

Vorbis is very simple because you really only have a quality setting to deal with, -q. If you want 128kbps, use -q4. If you want 192kbps, use -q6.

So you would enter at the command prompt:

oggenc -q4 myfile.wav

This will create a file called myfile.ogg that averages about 128kbps. There is a GUI front-end available called OggdropXP that allows you to encode files in single or batch mode.

Regarding the lack of bass, Vorbis has a known tendency to exaggerate the higher frequencies (at lower -q values anyway) so you might have gotten the impression that the bass was weak.

yup, lossy formats aren't cd quality. By definition, some data is lost. r3mix knows his stuff, but if he's still got that bit up about 256 kbps mp3's being "scientifically proven" to sound the same as cd-audio (wav files), he should d*** well remove it. Some people can tell a difference. Some can't. That said, ogg vorbis may be better than mp3. Though mp3 has been well-tuned at high bitrates and ogg still has a lot of room to improve - so in the future it'll definitely be better than mp3, which is close to the potential limits of the format. Mpc is the lossy codec that sounds the best to most people, and is definitely better than mp3 at higher bitrates. Good luck.

Which winamp version are you using? (2.7+ is good)

Are you using EQ? ("no" would be the right answer)

Are you using any effects plug-ins? ("no" would be the right answer)

Are you overclocking your machine? ("no" would be the right answer)

Have you enabled scaling, replay gain, normalizing etc? ("no" ... )

If all else fails, try decoding the mp3 to wav using lame, and compare the original and decoded wav files.

If you still hear a difference, could you post a short clip of the original .wav anywhere please?

Cheers,
David.

P.S. have you tried lame --alt-preset insane ?

The thread starter edited his post:

I think i found the answer now. Also I'm in the wrong topic sorry.
I use the equalizer to dampen bass because it's too much in my earphones.
If i deselect the equalizer i don't hear any difference in the sound.
But now it's harder to tell because theres so much bass.


This post has been edited by Alexander on Dec 12 2002 - 08:39 PM

FYI

Ehem...MPC --quality 8.5 --xlevel is for you then. It is one of MPC's suicide switches, (average 27x-320kbps) and it's called "BrainDead" for some reason .....It's even smaller than LAME --alt-preset insane, while it gives perfect quality even in transients!!.
I seriously you to switch to MPC.

Yes, seems that --alt-preset standard unfortunately fails now and then. You can very likely correct the situation by using Noise Shaping type 1.

That is:
--alt-preset standard -Z

unfortunately the bitrate is then also somewhat higher.

Hi.. Gregory here..

Have done extensive tests with LAME.exe 3.93 using the EXTERNAL selection and feeding switches to LAME. The exe runs in a machine-window in Win 2000 and even displays the VBR histogram. { love-it }

Here is a summary of the trials:


D:\Sade\Diamond Life\LAME trials

6,745,637 010 - Smooth Operator - V2 q2.mp3
6,762,320 011 - Smooth Operator - V1 q2.mp3
6,778,847 012 - Smooth Operator - V2 q2 - 128.mp3
6,817,056 013 - Smooth Operator - preset standard.mp3
6,834,307 014 - Smooth Operator - V2 q2 - 160.mp3
7,038,424 015 - Smooth Operator - V2 q1.mp3
7,053,847 016 - Smooth Operator - V1 q1.mp3
7,552,334 017 - Smooth Operator - V2 q2 - 192.mp3
8,885,551 018 - Smooth Operator - preset extreme.mp3
11,934,294 019 - Smooth Operator - preset insane.mp3

10 File(s) 76,402,617 bytes


After all this VBR encoding .. I'm hearing a `gritting' or `grating' in female vocals while playing MP3s cut with VBR as compared to the WAV file. I think it's harmonic or intermodulation distortion resulting from VBR (no data to prove this yet).

It is my observation at this point that a CBR setting of 256 kbps will produce a `clean enough' rendition. For all you guys that record noisy bands "banging on trash-can lids" (just joking) VBR will be good enough and save space .. but I don't think it's good for clean female vocals!!

Comments welcome...

Firstable, hopefully you didn't use 3.93 but 3.93.1. Since 3.93 was buggy.

Again, you should make sure that you are using Noise Shaping 1, otherwise such clean voices can be a bit of a problem. You can check this by using --verbose switch.
You can switch between NS1 and NS2 by using the -Z switch.

Thanks John..

Yes indeed.. I'm using 3.93.1 Lame.exe and LameEnc.dll dated 12-3-02.


Have looked through the docs (html) and can find no -Z switch or mention of NS. Do you mean the -X switch ?? Please advise on this feature!!


Cheers..!!

No, I mean the -Z switch.
If you do "lame --longehelp |more" in dos console window,you will see that -Z has been explained as "toggles the scalefac feature on".

Actually it's a wrong explanation... -Z switches between two "noise shaping" modes. Noise Shaping 1 and 2.
If you are using nspsytune (--nspsytune), try adding -Z, it will change to NS1. If you are using Gpsycho, well that psychomodel is not recommended, but it uses NS1 by default.
You can check in dos console which NS-mode you are using, by adding --verbose.


For advanced users:
Actually I learned today something new about NS1 which I didn't know before. NS1 uses "lossless" scalefactor_scale. If NS1 is in use and Lame notices that all scalefactors are even numbered, it halves the values thus saving some bits in coding. This is quality wise "lossless" and is no different outcome than using no scalefac_scale at all (besides the saved space).

With NS2, as it has been explained before, when scalefac_scale is in use (pretty often usually), scalefactor values are doubled. Since scalefactors are used to control the reduction of quantization stepsizes (which are use to control how much noise per SFB is allowed), the range or dynamics is larger for the stepsize adjustment. With NS2, scalefac_scale should be used only when it's "safe". Obviously this is not always succeeding, and NS2 scalefac_scale can introduce even quite bad audible distortion sometimes, especially when using vbr-old nspsytune. Good example is the erhu clip.

oh boy do i feel like a dumbass!! i've been using the EQ for a long time thinking that it improves the sound quality. after turning it off my files sound perfect. thanks alot

Thanks aGaiN.. John..


It's still ambiguous to me.. please define clearly like I'm a laymen.

Please advise if this is correct: adding the -Z switch means NS2 and not using it means NS1 default??


I need some better docs on this .. and the issue in question is HD and IMD (H and IM distortion) while using VBR encoding .. especially for clean female vocals. Typical switch settings are -V1 -q1 and -V2 -q2 with -b 128 or -b 160.



Thanks...

No. -Z switches between NS1 and NS2. For the best quality (but bigger size in vbr) you should use NS1.

Gpsycho uses by default NS1.
NSpsytune uses NS2 by default, so you have to use -Z to switch to NS1.

if you use just commandlines like -V2 or -b128, you are using the inferior gpsycho psymodel.

Gregory, use --verbose when using LAME. In the console output, you'll see a section labelled "psychoacoustic" and the fifth line down says "noise shaping: [value]". This tells you which noise shaping mode LAME is using. If the value is "2" then add the -Z switch to make it "1". If it is already "1" you do not need to add the -Z switch (because adding -Z would give you NS2). It's poorly designed but that's what we have right now.

Great answers.

I conclude the following from the site (this is a great site ).

1. Never use the EQ or anything alike when comparing the mp3 with the wav.
The sound will be altered quite much so don't use it for high quality listening either.
That's probably an old fact here at this forum that only a newbie as me didn't know.

2. --alt-preset standard is best for 180-220 kbit/s for now.
-r3mix is an older tweak giving inferior results to the above.

3. Use Lame 3.90.2 since it gives best sound (at least proven) with --alt-preset standard.

Please tell me if any of this is wrong.

Now another related question.
What parameters correspond to the --alt-preset standard?
I searched a lot but didn't find anything.
At least at the r3mix.net site the parameters were given.
Copy and paste from the site follows.
"--r3mix -b112" is synonym for "--nspsytune --vbr-mtrh -V1 -mj -h -b96 --lowpass 19.5 --athtype 3 --ns-sfb21 2 -Z --scale 0.98 -X0"
Also every parameter is explained at the site which is nice and "user" friendly.

I used the parameters instead of --r3mix -b112 and took away the scaling --scale 0.98 since i thought it is much better to use the normalizing option in EAC.
Choosed 96kbit/s bit Rate and the high quality radio button option.
Doing this doesn't alter the above settings i believe.

I think i want to use the --alt-preset standard from now on.
How about the --alt-preset standard. Does it contain the scaling i don't want to use?


Just a comment to the --r3mix settings i have used.
The sound is great. To my surprise it works good with my CD player with mp3 playing possibility. It supports variable bit rate and 320 kbit/s. So now I've got a reason for using the mp3 format. The sound from the player is probably not as good as from a good HiFi equipment but a lot better than from my sound card. At least it's adapted to my headphones. Remember i have good headphones (read first message in this thread for info). The portable CD player is some cheap shit (quite ordinary one) though but the sound is quite OK if you don't compare with high quality players. I must say it's hard to hear any difference compared to when using CDs with it.

There might be (and i think there is an extremely small) hearable difference but it's really on the limit of being perceivable.
For high quality enthusiasts i would say its not good enough. But they probably don't use lossy encoding any ways. But when listening for leisure in longer periods and not trying to find artifacts it's working. I think nobody will say "hey this isn't really a CD" unless you tell'em. In a blind test with "normal" music it is probably hard to say one or the other sounds better to the listener as long as you don't switch to high end stuff.

/ Alexander

What does --alt-preset standard that no custom command line can do ?
Re: LAME + Dibrom's presets/switches + remarks

OK That explains why i can't find what the presets correspond to as "external" lame parameters.
I thought something like this was the case but now i know more.

The only thing that is not clear to me now is the scaling.
Should i use the preset which seems to use a somewhat intelligent scaler (normalizer) and not use the normalizing (which is not stupid) in EAC instead then?

Also if I use the --alt -preset standard will it be "destroyed" if I use EACs external compressor option? I just guess it will add -b96 and -h if I choose to use it as above and it might "destroy" the --alt -preset standard?

To be sure to get te best result I might have to just rip to wav file and then use razorlame or somthing to do the mp3 coding with the --alt-preset standard option?

Just trying to be smart but as always it ends up in shit for a newbie, sorry. If it wasn't for this forum I must say the "user friendliness" of lame sucks.


/ Alexander

When you use very good headphones and are somewhat sensitive to artefacts, you should really take a look at the MPC section in this forum. It's a great codec, unfortunately not used by many people "in the wild"

Download mppenc 1.14 and start with --standard --xlevel. If that wont to the trick try --xtreme --xlevel

I let someone else answering about scaling (as I'm not sure), and about MP3gain.

The EAC configuration wizard (in the EAC menu) can allow you to use Lame.exe as external encoder. Just unselect your drives in order to skip their configuration (if it is already done) and jump to the external encoder config.
After having selected your Lame.exe version 3.90.2, choose "use recommended settings". Don't worry about the r3mix.net mention, it will associate "--alt-preset 128" with the "low quality" choice and "--alt-preset standard" with the "high quality" one, automatically generating this config line :

%l--alt-preset 128%l%h--alt-preset standard%h %s %d

As long as the link I gave figures into no FAQ yet (exept the old one in r3mix.net), I don't consider it as "shit for a newbie"

I just found the answer i got from PIO2001 when reading around in the MP3 Tech section.
But you are so fast at answering.
It's great.
I also read somewhere in this site that there is no scaling used in the --alt -presets.
So i will use the normalizing as before.

Thanks a lot. Now it begins to rock n roll.

/ Alexander

1) the presets do not use scaling (except alt-preset 128 which does). It is assumed that most people will use mp3gain

2) do not normalize with EAC, use mp3gain. If you wanna know why, I'll explain it later.

OK Thank you for that tip.
I read some about it just now.
I understood it seems to be the best choice (using mp3gain).
But now it begins to be quite expertish. I didn't understand it all yet but i hopefully will later. The scaling seems to be done kind of afterwards or something.
And its kind of better because the psycoacoustics will alter the sound otherwise or something? You don't need to answer now.


/ Alexander

Now I made some listening test on a piece of music with a drum cymbal and later in the piece a distorted guitar (yeah it's like that on purpose from the artists on this piece). I wanted to test the high frequencies especially since I think that sounded quite good with the -r3mix options (sorry I know I shouldn't speak to much of r3mix here... )

Anyways I tested standard with --lowpass 20 for comparison and the normal standard and the -r3mix. The 20 kHz didn't do much difference. It's not really that I can hear the higher frequencies than set by the standard preset but the files get a little bigger and so i guess the time resolution (sound over time smearing) gets a little better. How you perceive the sound, especially the high frequencies improves. With the r3mix option THIS PARTICULAR song sounded a little better than with the standard preset I think. I guess it's because it sacrifices some lower frequency characteristics for high frequency and that's a good choice in this piece. Compared to the 20 kHz preset it was about the same. But again it's hard to say. The differences are extremely small. Anyway the low pass seems to be set where it should be for preset standard according to me because the difference when making it higher was near nothing. This is because of the time resolution problem (pre echo) i think.

I didn't like the sound too much so I continued the test using preset extreme and insane.
Now the files are bigger. Especially with insane (yeah size begins to be insane here too).
I think extreme sounds a little better than insane??
I don't know really why but that's my impression. It should rather be the opposite. So I conclude from this that with lame there is no use to utilize the insane setting for me. I liked the extreme setting quite much. There is probably extra data that doesn't do much for how the sound appears to me but it felt like the "time smearing" problem got a little better.

Probably it will not be better than this. The time smearing (i guess its what people call the pre echo) is the problem. Hard to do much about that with the lossy compressors used today I guess. Somewhere on an other forum I saw somebody asking if there is any lossy compressing format utilizing wavelets. And the answer was NO. The point of using wavelets is that they are made to compromise between the frequency and the time domain (time is related to phase domain), so information regarding both these aspects are included. Normal music notes is an example of a simple wavelet like representation. This is not easy to implement though I guess. Also the coding and decoding will probably be more complex and that is not popular of course.

So here there is a problem with the lossy formats. Looking at the frequency domain you have made more or less what's possible as far as I can see.

If you want the nearest possible to CD sound i suggest the extreme preset. Then some margin seems to be included. The time domain problem is not possible to get much better than this though as long as you stick to mp3. Some more improvements might be possible to get in resolution for the frequencies you hear if more of the masqued frequencies can be removed but now it's a hell of a lot of tuning needed to get improvements in nspsytune I fear.

OK this is all just what I think. I am probably quite wrong somewhere since I'm not an expert on this. Don't regard the text above as a fact rather give a comment to this please. I will leave it here for now.

/ Alexander

It's standard procedure on the boards here that when you make claims about quality, especially if you're going to go in depth, that you back them up with abx results and/or provide samples so that others can verify or refute your results.

Also, the bit about --r3mix encoding more high frequency content than --alt-preset standard is somewhat of a misconception (there's more to it than simply what the value of the lowpass is set at). This isn't necessarily true, and especially not always true where it really counts. Since it appears that you didn't do any real scientific or blind testing or anything like that, it could be (and probably is) likely that --r3mix sounding "better" was simply the placebo effect.

The same thing goes for your statement about extreme sounding better than insane. There's a reason why we do blind listening tests here...

Do yourself a favor and test that way next time, you might be surprised at your results

Yeah dibrom is right!

I trust him completely.
The more testing i have made as a Newbiee i have understood Dibrom knows a lot more than I ever will about this. And also what he says has never been proven wrong to me.
As i have pointed out I have not used the best possible equipment and neither have i done the tests serious enough. i have not made blind testing for instance.
I made some more test (again not as serious as they should be done to be put on this site).
Sorry.

Still I feel I must say what I concluded from this.
Everytime I listen to mp3 made with the standard preset i feel a lot is missing compared to the CD sound.
And I still think there is some high frequency missing and also detail. It might be something else though.
It sounds "like" if somebody have put a pillow in front the speakers (but i use headphones).
I must also say that r3mix does NOT sound better overall when listening. The low frequency part of the distorted guitar is an example I can give. It sound really bad with r3mix. But somehow it feels less like the high frequencies are missing. But this might be because there is poorer reproduction of lower frequencies or ......
Actually I don't now reaaly why if you want to now.

To me the preset standard setting sounds warmer (meaning somewhat attenuated high frequencies or alternatvely more as i feel a small gain on all but the high frequencies) compared to the CD sound. This actually sounds very nice and good. It's also according to me a reason why some High End people like vacuum tube amplifiers better (but this is also a very controversial subject). Don't forget the ear is far more sensitive in the "mid" frequency area so in a way the sound can be accepted as getting "better" this way.
Dibrom and the other experts here of course know this and alot more I just write this for other readers sake.

As with everything, preffering a vacuum tube or a solid state (transitor) ampifier is just a matter of taste like how you want to put your equalizer settings.

A serious listening test though is ment to reproduce the sound as good as possible.
I just dont like the preset standard sound. I guess it's because i listen like more than 100 times (not consequtively though) on a song from CD so they are somewhere in my backhead. And then when I listen to mp3 I feel like something is missing. When listening to the same song in CD and mp3 format right after each other this is certainly what i feel.

The differences between preset standard and with the lowpass set higher is as I have said earlier hard to hear if hearable at all (probably something I imagine). I must say that I think that if I would do a proper blind test I would probably not hear any difference. Also the file size difference is peanuts.
This again is a proof of that Dibrom has done right.

I personally think there is no way to get better sound with lame and also keeping the high compression rate the preset standard has. So the preset standard fills its purpose and is OK to me as an alternative if I would mainly use it like on the computer or any other low quality sound set.

Ok so now to the good part. I wanted more detail and "high frequency" feeling so I tried preset extreme and insane. I liked the sound. Now i don't get this detail and high frequency missing feeling. To me it was actually a difference to the CD but again I strongly guess this is imagined perception. I believe that I would not be able to say wich is the original or the CD in a blind test. Although they actually might sound different how can I tell for certain that this is from the CD or mp3? I don't think i could bet on it. I also imagined insane was even better than extreme this time. As you might remember I preferred the extreme preset over insane last time. This just proves what dibrom says that if you don't do a real blind test the result is somewhat wrong. Therefore the whole testing and everything I have written here should not be at this site because it's not serious enough. But I feel it should be somewere?!

So at last MP3 with lame is good enough for me but i will never use anything less than the extreme setting again.

Also for test reference I feel I must say that the song with the cymbal and distorted guitar sounds is "You Are Here" from a single with the same name made by a swedish group named Brainpool.

For the result regarding "missing detail and high frequency" all songs i have burnt to a mp3 CD that was made with the preset standard setting are affected. I have made all my listening test on my portable CD player that also is able to play mp3 files.

Ok thanks for me. I also thank Dibrom for having put time to answer me. I feel VERY HONORED.
/ Alexander

Anyway, your tests are not very reliable, even for yourself, unless done in a blind fashion. For that we have the abc/hr comparator from our fellow ff123 ( http://www.ff123.net ).

Yeah my tests are not worth much. Especially not here among serious pros.
I'll skip saying more crap now.

Thank you for the link KikeG. I'll check this out a bit.
You are really helpful on this site. Thanks a lot.

/ Alexander