Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[bg0]

First - is anyone else having issues with getting the normalization to work consistently in MJ 8?  For me, it does not seem to always work - I can bring up the wav file in an editor after ripping, burning, or converting and see that it is not normalized to 98% as I have set in the options.

Second - does anyone care?  I have mentioned this many times and have not gotten any response.  Maybe it is just my problem or does no one else use normalization?  Or maybe no one else looks at the resulting wav file and just assumes it's normalized?

Third - is this a taboo subject?  I responded to a post about normalization not working correctly yesterday being very excited to find someone else willing to discuss it.  However, I look today and not only is there no response from MJ, but the entire thread is comletely gone.  What's up with that?

Is normalization a dirty little word that no one talks about but everyone does?

[Qwackamaster]

What's the point of Normalization?  Call me stupid, but what does it do?

[Doof]

I think it's supposed to make songs all the same volume, but it does so by making all parts of the song the same volume. So you lose the loud and soft parts of a song.

I don't think it's a taboo subject at all. I just think that a lot of the people here either don't use it because they don't care to, don't use it because it "ruins" the songs, or just aren't having a problem with it.

I use it when I make mixed CD's and when I transfer files to my Nomad. It seems to work fine for the Nomad, but I haven't had much luck with it on CD's.

[Matt]

>So you lose the loud and soft parts of a song.

Doof, that's not quiet right.  Normalization uses a constant scalar for each song, so it won't affect the dyanmics at all.

Also, we found a bug in the normalizer that would cause it to not awlways normalize... it'll be fixed next build.

-Matt

[Doof]

Well, see, I don't know what I'm talking about.

So if I'm understanding you right, normalization will raise or lower the volume of all parts of the song equally?

That's not quite how I understood it after reading a comparison to it and Replay Gain on the RG site, but I'm often times confused about the science of sound.

What are the different % values used for? Like, what's the difference between nomalizing something to 98% as opposed to 100%?

[Matt]

Doof, I think you've got it now.

Just to be sure, here's a quick sample of how it works:

Sound is a bunch of numbers that must fall in a certain range. (depends on the bit-depth)

So, pretend numbers can range from -100 to 100 and there's a sound that looks like this:

45, 30, -56, -72, -10, 30, 55

When we normalize, we first find the biggest absolute value, which would be 72.  That means the current level is 72 / 100 or 72%

Then, to normalize to say 98%, we need to scale it so that the highest number we found reaches 98%.  That means 72 needs to go to 98, so we multiply each number by (98 / 72) or about 1.36.  If we normalized to 100%, we'd multiply each number by (100 / 72) instead.

Hope that makes sense...

-Matt

[larry]

Matt

Working with your example.  What would Replay Gain do to the numbers. What is the downside of Normalization?  Why are other techniques used?

[Doof]

That makes perfect sense. Thank you.

One more question, though. I've often read that normalizing to 100% is a bad thing, and that you'd be better off staying at 98%. Why is this?

Oh, and I lied... more than one question...

Does normalizing have an adverse affect on the file?

and

Does it work across a group of songs? So if I normalize 3 different songs to 98% (let's say one is really quiet, one is really loud, and the other is in the middle) will they all be the same volume? If so, how does this, effectively, differ from Replay Gain?

I'm guessing by your example that normalizing multiple songs can have varying results depending on the initial bit-depth. In the case of converting one format to another, I'm assuming that they'll be the same volume because they are all first decoded to wave files with the same bit depth, and then normalized. Is this correct?

[Matt]

>Working with your example. What would Replay Gain do to the numbers.

Peak-level normalization only looks at the absolute highest peak.  That means 10 minutes of super quiet music followed by 1 second of super loud may already have a 100% peak (from that 1 second), but it'll sound quiet almost the whole time.

So, peak level normalization makes each song as loud as it can be without any clipping (numbers trying to go over 100%), but it does not make different songs sound the same volume.

That's where replay gain comes in.  It gets a lot more fancy and looks at how loud something sounds (for how to do that, check out replaygain.org), instead of just the peak level.  In the end, you still just get a fraction to multiply all the numbers in the sound with, but different files get different fractions depending on how loud they sound.

>One more question, though. I've often read that normalizing to 100% is a bad thing, and that you'd be better off staying at 98%. Why is this?

I don't know why people say this... I think maybe really old or shotty hardware didn't like 100% normalized CD's.

>Downsides

Well, any time you do DSP, you up the amout of "noise" in the signal.  That's why I think it's best to store the original exactly and then do any DSP's at playback time. (then changes aren't permanent)  Otherwise you can't change your mind later...

[larry]

Matt
For the ReplayGain, you said, "In the end, you still just get a fraction to multiply all the numbers in the sound with"

You cannot get ONE number to multiply all the numbers with.  If you did this you would have the NORMALIATION scheme, NOT ReplayGain.  Consider your -100 to |PLS|100 example, but with mostly -90s and just a few |PLS|90s.  No way to muliply everthing by one number and avoid clipping.  You would need a variable (dynamic) multiplier.  Either ReplayGain would clip or have to be a variable(dynamic) multiplier scheme.  Do you know which it is?

[Matt]

Replay gain would clip (and it can in rare cases) UNLESS it turns everything else down instead of turning stuff up. (which is pretty much what it does)

[Nikolay]

Normalization should be fixed in next build.

Nikolay

[LCtheDJ]

How high to normalize - 98%? 100%?  My testing has shown it's better to use 92% to avoid clipping.  This is as applied to the wav before encoding.  My tests showed that when a wav with a peak of 93% was encoded by lame.exe, and decoded back to wav, the new wav now has a peak of 99 - 100%.  Too close to the clip point for my comfort!  This is just for mp3 encoding, I haven't tested other compression formats.

I now normalize my wav files to 89% before encoding, and then use MP3 Gain to do a Radio Normalize adjustment to 89dB to get my songs to have the same 'apparent' loudness and still avoid clipping.  MP3 Gain makes a change to the mp3 file so the player adjusts how loud it is played back - this is not dependent on a database table.  All parts of the song are adjusted by the same factor.

To get the quiet parts of a song pumped up and the too loud parts reduced in volume, you would use a compressor/limiter.  Some plugins that do this are: AudioStocker Pro, Rock Steady, and Octiv OctiMax. I can post links later if you want them.

[KingSparta]

i was thinking that 100% is best

[Scronch]

My football coach always said to give 110%...

[JimH]

Wouldn't clipping be a problem if you gave 110%?

[Scronch]

Ooh, you punster, you.

Love it.

[bg0]

>>Also, we found a bug in the normalizer that would cause it to not awlways normalize... it'll be fixed next build.

>>-Matt

>>Normalization should be fixed in next build.
>>Nikolay

Thanks Matt and Nikolay!!!

I'll be trying it out!  

BTW - I have read a lot of discussions on normalization on other sites (Lame sites especially) and am having second thoughts on whether I should have done it or not.  But I've already encoded 3000|PLS| songs so it's kind of a mute point now.

Thanks again for the response!

[Doof]

YM
moot point
HTH

[shdbcamping]

The easiest way to understand normalization is to think of the stereo equipment that is being used. New stuff is more adaptable, old stuff is not. If you normalize to 100% older equipment seems to distort at lower volume (amplification) levels.
   Try a few different levels with the equipment that you have. Use the one that works best for what you use. I use 95% and it works with all my varied audio euipment. My stuff is "newer".  100% breaks up at much lower "Listening" volume.
Hope this helps!
Sean

[KingSparta]

>>RE:Normalization - a taboo subject???

Sade - The Sweetest Taboo.mp3

Warning!: My Wife Does Not Like Sade

[Brook]

It appears from what I'm reading here that normalization of existing MP3s works by decoding and converting them to wavs, normalizing the wavs, and then re-encoding back into MP3.  Is that true?  If so, they you'll definitely lose information as every MP3 encode/decode gets you further away from the original source.  

I read somewhere that there is an MP3 normalizer that normalizes in the MP3 domain without converting to a wav.  Does anyone know anything about it?  Perhaps it could be adapted as a MJ plugin.

[ZRocker]

Brook,

mp3Trim and MP3Gain are what you are looking for...mp3Trim is the definitive tool for this...MP3Gain with ReplayGain support is good too.  mp3Trim PRO is awesome!!  mp3Trim will have ReplayGain support later this year.