Well i wanted to do a test to see exactly how much loss you would get from going in the following conversions:

wav--->mpc--->mp3
as opposed to
wav--->mp3

and i got some pretty weird results..
according to what i got. the mp3 that was encoded from wav--->mpc--->mp3 ended up being better then the wav--->mp3

yes.. i know it sounds crazy.. might even be impossible to comprehend
but hear me out

i used Rightmark Audio Analyser to analyse all my results and i used mpc -q8
and the encoded bitrate for the mp3 was 256kbit

this is the command i gave lame :
-b 256 -m s -h -p -V 4 -B 256 --resample 44.1 -q 0

for mpcbatchenc:
--quality 8 --xlevel

basically the process was this....
1.) convert the original wav to mpc
2.) convert the original wav to mp3
[tried to get the same bitrate out of them]
3.)convert the mpc to mp3 (technically i converted the mpc to wav.. and then to mp3 using the lame command line up there)

finally i investigated the frequencies, Total harmonic destortion and more..
well according to what i saw. from the graphs.. that is what i conveyed..

if i have come to a wrong conclusion.. then please tell me.. i want to know as well.. but i think i read the results correctly..

as for abxing i have still to do that.. its pretty late and im in a rush..
also.. if you need any other graphs.. tell me and i will post them.

the pictures are here:
they are in png format and are very small for you 56k people.. even in zip..

Graphs/Data Images

well goodbye everyone.. off to sleep.. tired

*Obviously* you came to the wrong conclusion. It's flat-out impossible that wav-->mpc-->mp3 could be better than wav-->mp3.

What you're probably seeing is that certain inaudible information was dumped in the conversion to MPC (and then even more was dumped in the encoding to MP3), which might result in a "cleaner" appearance on a graph of some sort.

What you should probably do is look at the original .wav, then look at the (decoded to .wav) MPC and decide for yourself. Since you know (I hope!) that encoding to MPC can't improve on the original .WAV, this will be self-evident.

MPC has negative loss It is theoretically possible for each codec to cancel out each other's losses up to certain degree, isn't it? What does the analyzer software exactly do? In what sense it considers MPC transcode better? Why don't you skip the mp3 encoding phase and compare the results? like pure WAV and WAV->MPC

Hmmm... Is ist just me or is something with that cmd.line wrong?

dev0

Hello, Vegeta, welcome on Hydrogenaudio



Yes, I think so. There are some basic mistakes in your experiment.



This is the main one.
First I don't know how you can setup this program to compare files, for example, the intermodulation distortion is nonsense : 99.9 %
But as a rule of thumb, graphs can't account for a codec quality. It's covered in the FAQ of the board (direct link http://www.hydrogenaudio.org/forums/index....=ST&f=15&t=7834 , and there are many more, since the question comes back regularly).



This is also a mistake. Why such a command line ? You won't get better quality than using the recommended presets (see FAQ too).



The first thing to ABX would be your original against an alt-preset standard MP3. If you can't tell the difference, you can happily give up the above line for inferior bitrate and superior quality.
Then you may try to ABX the direct MP3 against the transcoded one. If there is an audible difference, then the transcoding have decreased sound quality.

There is also a sticky "recommended MPC settings" thread in the MPC section of the board. Quality 8 seems overkill to me.

EDIT : by the way, Dev0, answering this kind of posts, it would be kind of you to point users to the recommended Lame settings, or to advise the alt-presets.
If Vegeta uses that line it's because he thinks it's good.

Yeah, it's a little creative.

ok everyone thanks for the criticism,
ill read over what you guys said carefully. and perform a better experiment this time around

like i said, i know there is a great possibility i am wrong, its pretty hard for even i to believe that such a conclusion could be true. i just wanted to see if someone could point out why i was wrong.. thank you ..

i now get the little joke about the the variable bit rate lame command line which is really a 256kbit CBR.. heh. well i didnt have so much time to read last night.. i was in a bit of a hurry. i will be using the recommended settings for both encoders this time though..

I have to admit, that this was not very constructive at all. I'm sorry.

The quality degradation of transcoding from MPC to MP3 has been discussed quite a few times here for various reasons (portable support etc.) and the preliminary conclusion has been that the degradation is not quite as big as the one from MP3 or Vorbis since Musepack is a subband-codec, while Vorbis, MP3 and AAC are transform-codecs.

dev0

If i'm not mistaken, it was all about the current implementations of transform codecs.

I don't think it is correct to generalize it to transform coding as a technology.

ok i performed the same tests. only i used the following command line arguments for the two encoders:

lame:
--alt-preset standard

mpcenc:
--quality 6 --xlevel

i wont say any conclusion right now.. [i would hate to be wrong more then once]
so here are the results.. tell me what you guys think.. according to the graphs.. and the graphs alone.. what does it tell you..

besides.. if any of you would like to do the test i am doing.. you can download the Rightmark audio analyser from their site.. do a search on google it will come up..

last but not least for those of you say that GRAPHS donT ALWAYS say which is quality is better i think you are wrong.. in some situations they do.. and clearly this is one of them. look at the graph of the mpc vs. the mp3
im sure you will clearly tell which codec is better.

as for the abx test.. just did that.. for all i know i cant distinguish between those two files.. (the wav->mp3 and the wav->mpc->mp3)
not one single bit..

Graphs/Images 2

can't see the graphs:

ERROR:

with IE and Mozilla Firebird

Vegeta> Graphs don't always give perfect results about hearing quality... Man doesn't hear the music graphs but the music itself. Human is not a machine and it's not perfectly hearing everything, that's why psychoacoustic models exists...

amano.. sorry for the error. i am personally running the file now.. via ftp..so load up your favorite download manager (personally flashget will do )

yes i understand Graphs dont necessarily show whats best all the time. but in this case i think it does.. in the abx test once again i mention.. i wasnt able to distinguish between the wav->mp3 and the wav->mpc->mp3..

If you can't ABX the difference we don't really care at all.
First you ABX to see if there's an audible difference; then you can use graphs as a tool to find what's wrong.
How the graphs looks are irrelevant and uninteresting if you can't hear (ABX!) the difference.

im not trying to prove anything relating to what you mention Jan S.
im trying to prove that going from wav->mpc->mp3 isnt really AS HORRIBle as some people make it to be.. i just thought i would help by actually testing this out..

basically im saying.. transcoding from mpc->mp3 wont degrade the FILE enough to be able to notice greatly between the wav->mp3

Point taken : you can't hear a difference. Thank you for testing. What was the hardware and music used, by the way ?



The frequency response graphs tells you how the different codecs lowpass the sound, they don't give the slightest information about quality. You can encode at very low bitrates and keep all frequencies with the appropriate switches, this sound horrible.
The thread I gave you above is exactly about this : a worse setting giving a better graph. What did you think of it ?

PIO.. hmm you are right about that.. but regardless. the fact that they at least to me seem transparent.. is ok enough.

question.. is there any way to actually test the quality without some sort of audio analysis? [not listening abx tests]

as for hardware.. i have a Turtle Beach Santa Cruz Soundcard and the song i used was Blink 182 - Everytime I Look For You [ROCK]

If a graph told audio quality, then we all would be using BladeEnc or Xing(old)

Unfortunately, no. There have been some programs around simulating the ear's response to psychoacoustic encoders, but if I'm not mistaken they were only useful for low bitrate analysis, (or not at all ?).

The main problem was summarized by Gabriel not long ago ( http://www.hydrogenaudio.org/forums/index....25&#entry116011 ) :



That's the problem of high quality encoding. We're looking for perfect encoders. If we could know the quality of an encoded file with a given method, this method would be used to make a perfect encoder !

hmm i see..
well hope i didnt stir any trouble around here.. bUt i want an answer to all this..

guess i will have to keep waiting..
besides the point at least i feel somewhat more comfortable transcoding..

The program you refer is called http://www.hydrogenaudio.org/forums/index....highlite=EAQUAL.