Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Not Including --scale? (Read 5993 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Not Including --scale?

I know it's better to use mp3gain than --scale, but is it such a good idea not to include --scale in the mass market --alt-presets? People that only know about the alt-presets but not mp3gain may think there l33t 3ars can still hear lots of problems with --aps and declare that MD is better, or something, when all they're hearing is just clipping. Besides, has anyone bothered to try ABXing a mp3gain'ed mp3 from one that has been --scale'd to the same level?

Not Including --scale?

Reply #1
--scale is only included in ABR/CBR presets not in the High Quality VBR ones. The assumption made was, that people who'd use CBR cmd.lines wouldn't dare to mp3gain there files anyway.

dev0
"To understand me, you'll have to swallow a world." Or maybe your words.

Not Including --scale?

Reply #2
Quote
People that only know about the alt-presets but not mp3gain may think there l33t 3ars can still hear lots of problems with --aps and declare that MD is better, or something, when all they're hearing is just clipping.

With too much scale you would have people complaining about loss of bass punch instead. Choose your poison.

Not Including --scale?

Reply #3
How about a default of --scale 0.95

That's not 'too much' is it?

And why would it only be the bass that's affected?

I need to check to make sure, but I think I put some un-mp3gained files in my last mp3 CD by accident and they don't sound good...

Not Including --scale?

Reply #4
Scaling has too many downsides. It's not lossless, which equals a quality reduction (especially troublesome for those that use MP3Gain anyway). It slows down encoding. It doesn't prevent clipping per se, since it doesn't examine the file to find out the necessary amount of scaling.

Also, we would surely give the wrong signal to all alt-preset users. Many would figure that scaling must be a good way to get rid of clipping, and therefore dismiss MP3Gain. We should rather promote MP3Gain more.

Not Including --scale?

Reply #5
Quote
Scaling has too many downsides. It's not lossless, which equals a quality reduction

Oh, come on! There is (should be) no quality reduction implied when you scale your input. Please provide some blind tests before stating such things. ABX the difference between an encode using --scale and another one that is mp3gained to the same level, but otherwise with the same settings. Good luck!

But there are other reasons why I feel it's better to leave the scaling/replaygaining to the user. (See above)

<edit>I remember LAME had problems with some low level clips, but IMO that's a reason for doing normalisation (replaygain is one form of normalisation) before encoding rather than after.</edit>

@Joe Bloggs
The reason I mentioned bass punch in particular is that it is the complaint I have seen most often. For some reason it seems like people love booming bass and if you lower the volume for them it's the bass loss they will notice first.

Not Including --scale?

Reply #6
Quote
Oh, come on! There is no quality reduction implied when you scale your input. Please provide some blind tests before stating such things.

Did i say "blatantly audible quality reduction"? Although the quality reduction is mathematically provable, all i wanted to point out is that there are so many lossy stages the audio went through (even way before the CD has been pressed), why would we want to perform such unnecessary steps like scaling? There is just no benefit in this.

Not Including --scale?

Reply #7
Comment on your <edit>: No, ReplayGain is not a form of normalization.

Not Including --scale?

Reply #8
If it's not audible quality reduction then why should it be an argument against using --scale? And the scaling in LAME is implemented with floating point type (at least 32 bit) so the quantisation error is really neglectable.

Is Replaygain not a normalisation of the perceived loudness? Then what would you call it?

Not Including --scale?

Reply #9
Oh, and by the way:

- Scaling does not slow down encoding (at least not in my very limited test)

- Scaling by a value less than 1 reduce the number of clippings. But as you say it does not prevent all clippings every time.


My point: There are other reasons why --scale should not be included in a standard preset in LAME, but I can't see the logic behind those you listed. Sorry.

Not Including --scale?

Reply #10
Quote
If it's not audible quality reduction then why should it be an argument against using --scale? And the scaling in LAME is implemented with floating point type (at least 32 bit) so the quantisation error is really neglectable.

Well, there are enough arguments against a fixed --scale value in the alt-presets, this is just one of them.

Quote
Is Replaygain not a normalisation of the perceived loudness?


I see there's a diversity in interpretation here. When you mentioned normalization, i thought of the process of normalization (which involves requantisation), not the end result.


Quote
Scaling does not slow down encoding (at least not in my very limited test)


Oh yes, it does. Ask my brother, he has a Celeron 300.

Not Including --scale?

Reply #11
Nice pictures, but are you sure that you used the same float type as LAME to represent the samples? It lookes like a lot of noise for 32 bit precision..

You are right about the speed difference. I didn't test that properly.

Not Including --scale?

Reply #12
Quote
Nice pictures, but are you sure that you used the same float type as LAME to represent the samples?

No, since i was talking about normalization and the difference to ReplayGain.

Not Including --scale?

Reply #13
Like the one used in EAC and other cd rippers (using 16 bit fixed point)? Then of course there will be noticable quantisation noise if you repeat it enough times. I thought we were discussing the --scale function of lame...

Not Including --scale?

Reply #14
Quote
I thought we were discussing the --scale function of lame...

But hey, you were the one who brought this up ("replaygain is one form of normalisation", from my point of view incorrect). 

If LAME really uses 32 bit floating point, then you're right, it should not be a decisive factor in this discussion. However, due to the other points mentioned, i'm still against scaling in the presets.

 

Not Including --scale?

Reply #15
Scale is not included in the vbr presets, because vbr in LAME doesn't have the horrible problems with clipping that abr does, especially as bitrate decreases.

Beyond that, scaling at a fixed amount makes no sense because all albums are mastered differently, and different genres of music would typically require different levels of scaling.  Someone who listens to well mastered classical music is probably not going to want their tracks scaled for example.  Someone who listens to pop music mastered for the radio though may end up needing to scale by huge amounts to prevent clipping.  One size definitely does not fit all well enough in this case that it'd make sense to include a default scale value in the high quality vbr presets.

The only reason the abr include a fixed scale value is because this is the value that the majority of severely audible clipping seemed to go away on most of the samples tested.

Not Including --scale?

Reply #16
I have this vague feeling that I have gone through all this before... sorry guys :x

Not Including --scale?

Reply #17
Quote
- Scaling by a value less than 1 reduce the number of clippings. But as you say it does not prevent all clippings every time.

And mp3gain prevents this all the time.

replaygain is not a form of normalization
but normalization is a form of replaygain

Not Including --scale?

Reply #18
Quote
And mp3gain prevents this all the time.

Of course that depends on how you use it. Track or Album mode of operation in mp3gain doesn't prevent clipping. And I'd say those two are the most frequently used modes. But this is just off-topic theoretic speculation anyway. The original poster's question has been answered.

Quote
replaygain is not a form of normalization
but normalization is a form of replaygain


I guess I have a little bit wider definition of normalisation.