how to use azid 2pass

[PeterTheMaster]

whats the command line switch if i want azid to find the maximum gain and then apply it?
this is better than converting to wav and then normalize, right? because of the float input that azid can still use but normalize cant?

[PeterTheMaster]

maybe im wrong but i think i have heard that lame can handle float input.
so is azid able to produce float output, if yes, how? (commandline wise)

because this way quality should be a little bit (maybe unhearable) better in the end, right?

[DJ Bobo]

Typical command line:

azid -a -c normal audio.ac3 audio.wav

The "-a" switches to automatic 2-pass mode with automatic gain control

[PeterTheMaster]

so mine would be: -a -cheavy -L-3db "eng.ac3" "eng.wav"
is it important where to put that -a?
becuse i want to be sure that normalization takes place after drc.

[DJ Bobo]

Sorry, I don't know, I use usually the above mentionned command line.

If I'm in a hurry, I make one pass AZID decoding without normalizing, I note the peak value, encode directly to mp3, then use MP3GAIN or MP3TRIM to amplify with the corresponding value (that goes extremely fast!)

[DSPguru]

you can use BeSweet's post-gain engine.
it decodes the AC3 using azid.dll, and encodes to MP3 using lame_enc.dll, while it does that it finds the maximum gain, and then it increases the volume in a similar way to MP3GAIN.

@bobotns
BeSweet is faster and will acheive better quality than your alternative, 'cause all arithmethics inside BeSweet is 32bit floating-point, while in your suggestions it's only 16bit integers.

[PeterTheMaster]

azid.exe -a -cheavy -L-3db "eng.ac3" "eng n.wav"
lame.exe --alt-preset 128 "eng n.wav" "eng n.mp3"

this way sounds really cool, so what would be the besweet command line parameters for the above process?

and i havent heard much about normalizing an mp3 file before, but shouldnt this be the most unaccurate way to do it?

[DSPguru]

one pass normalization :
BeSweet.exe -core( -input "enc.ac3" -output "eng n.mp3" ) -azid( -cheavy -L-3db ) -lame( --alt-preset 128 ) -ota( -G max )

two-pass normalization :
BeSweet.exe -core( -input "enc.ac3" -output "eng n.mp3" ) -azid( -cheavy -L-3db -g max ) -lame( --alt-preset 128 )

make sure to update to the latest BeSweet & BeSweetGUI beta releases.
also - go read the BeSweetGUI guide !

[PeterTheMaster]

i do not want to do it with a gui.
but now i see how i can give any lame or azid parameters, you found a nice way to handle this.

which way gives (in theory) better quality?

do i have to update the lame and azid dlls myself or do you make a new besweet release everytime lame or azid is updated?

[DSPguru]

Originally posted by PeterTheMaster
which way gives (in theory) better quality?i say two-pass, others say one-pass. do a listening test and judge for yourself.

do i have to update the lame and azid dlls myself or do you make a new besweet release everytime lame or azid is updated? just download the latest official BeSweet release (v1.2) and overwrite with the files supplied with the latest beta (v1.3b6). the azid.dll (v1.8) is standard, but the lame_enc.dll is something i made, so you can't find it elsewhere.

[PeterTheMaster]

i think this shouldnt be a matter of listening test.
the normalization in azid multiplies floats, right? so this should be more accurate than working with the rounded and processed values in the mp3.

while im at it:
1. how do you make the post gain? on the mp3gain page i read that it doesnt gain to max but analyses how loud it sounds. whats that?
2. lame can process mp2 files too. how would the commandline you gave me above look like if i would use mp2 for input instead of ac3?
is post gain available for mp2 conversion?

[DJ Bobo]

Originally posted by DSPguru
@bobotns
BeSweet is faster and will acheive better quality than your alternative, 'cause all arithmethics inside BeSweet is 32bit floating-point, while in your suggestions it's only 16bit integers.

I know, I talked with Avenger about it and he proved me that on a weird sound sample, but this is only theory, in practice I hear no difference between the 16-bit method and the 32-bit method.

[tangent]

Originally posted by PeterTheMaster
azid.exe -a -cheavy -L-3db "eng.ac3" "eng n.wav"
lame.exe --alt-preset 128 "eng n.wav" "eng n.mp3"

this way sounds really cool, so what would be the besweet command line parameters for the above process?

and i havent heard much about normalizing an mp3 file before, but shouldnt this be the most unaccurate way to do it?


If you intend to use mp3gain or postgain, make sure you use "--scale 1" after "--alt-preset xxx". e.g. "--alt-preest 128 --scale 1". This will stop LAME from pregain scaling (defaulted in the --alt-preset)

@DSPGuru
One thing I forgot to check before, with BeSweet using the --alt-preset, did you make sure the "--scale 1" goes in?

[tangent]

Originally posted by PeterTheMaster
i think this shouldnt be a matter of listening test.
the normalization in azid multiplies floats, right? so this should be more accurate than working with the rounded and processed values in the mp3.

while im at it:
1. how do you make the post gain? on the mp3gain page i read that it doesnt gain to max but analyses how loud it sounds. whats that?
2. lame can process mp2 files too. how would the commandline you gave me above look like if i would use mp2 for input instead of ac3?
is post gain available for mp2 conversion?

Theoretically, using floats will provide more accuracy and avoid rounding errors. In practice, no one hear these errors, and when encoded to MP3, they fall below the ATH curve anyway.

MP3gain can do gaining to max (although that's not the default behaviour).

[PeterTheMaster]

hm, sounds you know mp3gain. do you also know how to control it via command line?

[tangent]

Not yet. Haven't tried.

[DSPguru]

@bobotns
pay attention, the differenece isn't between 16bit integer to 32bit integer (*).
in case you're doing a 2pass azid for normalization, the difference between 16bit integer to 32bit float is equivalent to (*).
but, if you want to implement a post gain, you'll probably loose most of the signal information by working with 16bit integers, while in 32bit floats the signal won't be harmed.

@tangent
my dll implements the official alt-presets, meaning scale in this specifiec case is 0.93.
what i could do (& did) is offer access to the --scale switch.
so it's now supported in v1.3b7.

Dg.

[PeterTheMaster]

how can mp3gain do gaining to the max? i found it neither on their page nor in the docu.

[DSPguru]

mp3gain doesn't gain to the max, it gains to replaygain : http://privatewww.essex.ac.uk/~djmrob/replaygain/calculating_rg.html.

replygain is a great concept, but i find it irrelevant to divx/xvid encodes.

[DSPguru]

Originally posted by PeterTheMaster
how can mp3gain do gaining to the max? i found it neither on their page nor in the docu. Peter, you are the laziest butt in this forum !!
Originally posted on http://www.geocities.com/mp3gain/
MP3Gain analyzes and adjusts mp3 files so that they have the same volume.
It does not just do peak normalization, as many normalizers do. Instead, it does some statistical analysis (http://www.replaygain.org/) to determine how loud the file actually sounds to the human ear.do you really think that distinguished users like tangent should waste his time and and explain here how mp3gain works, or how to execute mp3gain via cli, just because you're too lazy to follow a link ?!
you're the FIRST user that asked me such a trivial question like - how do i convert from ac3 to mp3 using BeSweet's cli. why do i spend my time on creating the 'help' directory in BeSweet package if ppl like you are too lazy to read ?!

i'm telling you for the last time, Peter, stop being lazy, or you just won't be.

[PeterTheMaster]

i have read what you quoted on the page before but it didnt help me.
tangent wrote a few posts above:

MP3gain can do gaining to max (although that's not the default behaviour).

so i asked how since i didnt find an explanation anywhere.

about besweet: the only thing i found on your page was a link to
http://www.doom9.org/ac3tomp3.htm
which only describes the gui not the cli parameters.
i had not downloaded besweet then. sorry.

does the new version now support alt-preset for abr?

[DSPguru]

Originally posted by PeterTheMaster
i have read what you quoted on the page before but it didnt help me.
tangent wrote a few posts above:

MP3gain can do gaining to max (although that's not the default behaviour).did you follow the link ?! the explenation is THERE !

so i asked how since i didnt find an explanation anywhere.Anywhere ?! i gave you a DIRECT link :
Originally posted by DSPguru
mp3gain doesn't gain to the max, it gains to replaygain : http://privatewww.essex.ac.uk/~djmrob/replaygain/calculating_rg.html.

-
about besweet: the only thing i found on your page was a link to
http://www.doom9.org/ac3tomp3.htm
which only describes the gui not the cli parameters. CLI refernce can be found here (http://dspguru.notrace.dk/cli.html).
i had not downloaded besweet then. sorry.you didn't download the progarm, you didn't try to work with it, you didn't read all its' attached 'help' directory, and you come here for support ?! that's an exact violation of our rulez ! here's the right sequence :
a. play with the program, try figure things by yourself
b. if you fail, read guides, faq, and search for old posts
c. if you still fail, start a thread.


does the new version now support alt-preset for abr? i'm amazed ! are you sure you visited the BeSweet page ?!

[PeterTheMaster]

ok, its in the beta7, i just came from reading the help directory of 1.2 ...

but i read everything on the mp3gain site and didnt find how to gain to max. the only way i can fiqure out is to try until there is no more clipping.
if nothing more is possible why cant you just say it?

[DJ Bobo]

Originally posted by DSPguru
@bobotns
pay attention, the differenece isn't between 16bit integer to 32bit integer (*).
in case you're doing a 2pass azid for normalization, the difference between 16bit integer to 32bit float is equivalent to (*).
but, if you want to implement a post gain, you'll probably loose most of the signal information by working with 16bit integers, while in 32bit floats the signal won't be harmed.

I know, but still there is no difference in my ears, this is all theory and audiophile stuff what you're talking about.
I've compared AZID normalization with post normalization in MP3TRIM, both files sound _EXACTLY_ the same on my SONY mini system.
Avenger sent me a weird AC3 file which level he reduced to -75db and told me to normalize it, in that case there was a BIG difference. But this is a very special case, in practice there isn't such cases, so I won't bother my self and waste 15 minutes to do normalizing in AZID instead of MP3TRIM.

[DSPguru]

@bobotns
i didn't say "go do a two-pass normalization", i agree that one-pass normalization is very good, i just suggest that you try doing it with BeSweet (instead of azid,lame,mp3trim) 'cause :
a. it's faster
b. quality is better (32bit floating-point).
c. no intermediate 2gb wave file.
d. you have on-the-fly sample-rate-conversion & delay assertion.
e. you have 4 different DRC algorithms.
f. it can handle list files.
g. it will create a useful logfile with all info, when finished.
h. it can add a wav header to the mp3 for easy muxing.

[PeterTheMaster]

so would you please prove my stupidity by quoting the page on how to gain to the max?
and if it is not possible: i didnt suggest its possible in the first place, it was tangent.

[DSPguru]

Originally posted by DSPguru
mp3gain doesn't gain to the max, it gains to replaygain : http://privatewww.essex.ac.uk/~djmrob/replaygain/calculating_rg.html.

[PeterTheMaster]

so you say its not possible and tangent was wrong?
i dont see how that make ME stupid.

[DSPguru]

a. mp3gain can apply a constant gain. that could also be the maximal value.
b. i'm not interested to prove you're stupid. i wish all my users here would be smart and unlazy.
c. if you read the audio faq, q14, you'd know about mp3trim, that DOES offer max gain assertion.
d. did you give a try to the command-line i gave you yesterday :
Originally posted by DSPguru
one pass normalization :
BeSweet.exe -core( -input "enc.ac3" -output "eng n.mp3" ) -azid( -cheavy -L-3db ) -lame( --alt-preset 128 ) -ota( -G max )now, what's there that you're missing with this solution ?

[DSPguru]

Originally posted by PeterTheMaster
ok, its in the beta7, i just came from reading the help directory of 1.2 ...did you read line 12 in here (http://forum.doom9.org/showthread.php?s=&threadid=15738) ?

and what about the quote from my reply yesterday (http://forum.doom9.org/showthread.php?s=&postid=110259#post110259) :
Originally posted by DSPguru
make sure to update to the latest BeSweet & BeSweetGUI beta releases.

[PeterTheMaster]

i did not read this point 12 because i was not going to post any bugs.

i had no possibility yet to test your command line, but i already updated my .bat files that i use for audio conversion. but i didnt chose this line but the one with 2pass azid.
but now i am almost convinced to switch to ogg, but before asking if your tool supports this i will go reading...

thanks for your help and being so patient.

[DSPguru]

Originally posted by PeterTheMaster
but now i am almost convinced to switch to ogg, but before asking if your tool supports this i will go reading...Thanks, Peter :D.

btw, one-pass-normalization is only possible with MP3.

Dg.

[tangent]

MP3gain does not really implement the ReplayGain standard (simply because there is no support for it). It uses the techniques outlined in the document to calculate the "average volume" of the clip. It tries to implement ReplayGain in a different way using the GlobalGain field to allow users to keep volume across many clips to be the same (actually within 1.5dB which is the limitation of the GG field). This is the primary purpose in which MP3gain was designed and made.

A few of us then decided that the GG field is also a great way of normalizing to max, so due to popular demand, the author put in features to allow max noclip gain with mp3gain. To use this feature in the GUI, you have to Options->Advanced and check "Enable 'Maximizing' Features". Open all the mp3 files you want to maximize, then from the pulldown menus (max noclip features not available in the buttson yet), select Max Noclip Analysis. After the analysis, select Apply Max Noclip Gain.

[tangent]

Originally posted by DSPguru
@tangent
my dll implements the official alt-presets, meaning scale in this specifiec case is 0.93.
what i could do (& did) is offer access to the --scale switch.
so it's now supported in v1.3b7.

Personally, I think you should by default override the --alt-preset's --scale. If a user uses pre-encode normalisation, then it would be silly to do another --scale in LAME. If a user uses post-gain, then --scale is useless. Either way, --scale should not be used.

[DSPguru]

-lame( --scale 1 ) should be used along with -ota( -G max ).
as for -azid( -g max ), you're right, it shouldn't matter.