Does anyone noticed that the "--scale 0.98" in the r3mix setting does not work at all??
Only if you go as low as 0.80 can notice that the peak value is normallized to that value.
Also, when using EAC + r3mix setting I sadly discovered that the normalization option is kinda fucka messin'!
My choice usually was 98% normalizing and then mp3 encoding with the r3mix setting.
Usually RMS and peak values for the decoded mp3 and the original wavs were matching (as it should be); But ..
...but if your wav's (CD) are already at high level (say f.e. 98 - 99%) the the LAME (r3mix, VBR, ABR, whatever) codec take this level even a little higher and then CLIPPING OCCURS!!!
If the CD is moving under these peak values (say less than 85%), then original and decoded files' peak match as I said before
WHY????
There's no point in doing "automatic ripping" if you must check if clipping occurs for each file.
DOES ANYONE NOTICED THIS??
Thanks
MP3 always changes the precision of certain frequencies and discards others. When changing the frequency composition of a waveform, the peak values may change up or down (they will often go up even if you remove frequencies).
The idea of lossy coding (like MP3) isn't to reproduce the exact same wave shape, but to make it sound the same to the human ear, so it may have very different peak values to the original CD waveform.
For this reason r3mix tried to scale the amplitude to reduce clipping, but 98% isn't enough for ultra-loud pop music of today (a couple of years after r3mix stopped), though clipping might not be audible.
Note that --r3mix is pretty good, but --alt-preset standard is even better at handling 'problem samples' and is really highly tuned with blind listening. It's only slightly large file size too.
If you're prepared to leave behind MP3, Musepack (.MPC) is even better in quality and file size (but not compatible with portable players).
Most CDs today are mastered much too loud anyway, so many of us with collections of music on our computers use ReplayGain to correct for this and allow us to comfortably shuffle old and new music.
For MP3 files, mp3gain is the program to use. This decides how loud the music sounds to the human ear, rather than measuring the peak value, and allows you to adjust on a track-by-track, or album-by-album basis (I like the latter).
Doing "Album Gain" you can analyze a whole ripped CD and find the peak values plus how loud it sounds as a whole. It will suggest the amount of correction needed to make its loudness 89 dB and can apply the changes without any loss of quality.
So you only need to run MP3gain after you've ripped a CD using EAC and it will warn you in red if any clipping would occur with either Radio Gain or Album Gain applied.
That's really easy, and mp3gain can run while you get on with ripping your next album.
My preferred solution for MP3s:
• EAC (secure mode) ripping then encoding with
• Lame --alt-preset standard or...
• Lame --alt-preset standard --nogap (for attempted gapless encoding)
• MP3gain to 89 dB volume in Album Gain mode.
• PC Playback in WinAmp 2.81 using mp3splice output plugin to remove gaps perfectly in live/mix albums, even if you didn't encode with --nogap.
Album Gain mode is good enough for shuffle playing of mp3s, so I'm OK with that.
If I'm not tied to .MP3, I prefer to replace lame --alt-preset standard with 'mppenc' (default option is standard) which results in .MPC files that sound just as good as the CD or lame APS (but problem samples sound better than lame) and, being more efficient, saves a little disk space too!
.MPC files also play back just fine through the mp3splice plugin, and replaygain is stored in tags. EAC might support replaygain scanning in future, but for now, I use the Foobar2000 player to apply ReplayGain info to .MPC tags. I can then choose to have Radio or Album Gain at the time I play back the files.
Regards,
Dick Darlington
Two small corrections (call it nitpicking if you like ):
bROTHER wrote:
Only if you go as low as 0.80 can notice that the peak value is normallized to that value.
The --scale option does not describe which peak level the file gets normalized to, but simply by which factor the input gets scaled. I.e., if you specify 0.80, the amplitude will be scaled down by 20%.
DickD wrote:
.MPC files also play back just fine through the mp3splice plugin [...].
Actually, they will also play gapless if you just set the "buffer-ahead on track change" option in the WaveOut or DirectSound plugin - no need to "hack" the gaps out of them like you have to do with MP3s, because there are none!
DickD,
many thanks for your exhaustive reply.
Yep, I see the logical on that. But, concerning to Lame settings, I'm stubborn to use CBR or ABR presets. maybe the r3mix set is from years ago, but blind listenings from then are still up to date by definition.
I think the VBR algoritms (the right ones) are still more advanced.
We're seeing a handful of new formats based on psychoacustics poping up everywhere, but their only advantage can be file size, cos quality is just reached. This file size isn't as large as to change inmediatly to the new formats, cos years of fine tunning should weight more. Of couse there are other important technical details such us gapless encoding, extended tagging, etc. But tunning is needed for all of them plus a "worlwide" knowledge (in P2P file sharing systems is hard to find the quite old ogg files).
But, as said before, from the sound quality point of view, they are all the same (though what some people say over the net).
Also, compatibility issues are important. I own a SlimX mp3player which at the moment only decodes mp3 and wma.
Well, I've just downloaded mp3gain and need the time to give it a try. And also de -nogap setting.
C ya!
Best regards,
©bROTHER
Hi again,
I've just gave it a try (mp3gain), and ... well ... not bad ... But for most albums (Pop, dance, rock) I prefer max gain without clipping. After all, these albums ara composed of singles plus somthing else (right?)
Only in special albums like live recordings where songs are chained, or with special tracks (like vocal ones) or classic music I think is useful a "per album gain", and of course using the max gain (without clipping) available. In this way I'm boosting up the SNR but keeping the compositor intended effect cos preserving the differences between tracks.
By the way, this brings a question to my head (taking again the -scale 0.98 set): making everything as I stated in the first post (i.e. wav 98% norm + mp3 -scale 0.98), are the mp3 really getting clipped or it's just a matter of the frame gain? I think the file is really getting clipped forever, even if we reduce the frame gain afterwards ...
But if the input get's scaled as Volcano stated (danke schon für deine post), then the codec itself is reading a non-clipped file and then isn't trying to encode the odd armonics that appear when clipping occurs (which normally traduces in bigger files cos more frecuencies are present), so no reason to think that the encoded file is clipped, just a matter of changing the frame gain, huh?
Bye
Mach's gut!
©bROTHER
I think the file is really getting clipped forever, even if we reduce the frame gain afterwards ...
No, this is not so. The encoder does actually encode the samples above full-scale, they will not be clipped off until you decode (=playback) the file. This is why MP3Gain can totally remove clipping that was introduced by the MP3 encoding process, and why scaling before encoding or anything like that is not required.
OK, thankx
And, what can you say about the statement "normalization introduces rounding errors UNLESS YOU RE-DITHER"?
ciao