Topic: normalizing before or after tracking

[SparkE!]

Brian's right about this.  Normalizing amplifies everything by the same gain factor and sets the loudest sample of the original recording to 0 dB signal level (the maximum possible).  If you want to bring up the quiet parts so that they are not so much lower in volume than the loud parts, then you use compression to intentionally reduce the dynamic range in the recording.

In my opinion, this is the procedure to follow if you are going to use compression:

First convert it to a floating point representation with the highest numerical resolution possible (usually 32-bit floating point), then normalize it, then compress it and re-normalize it.  You will want to normalize it first because you want the loud parts to already be close to clipping.  That way, when you compress it, the loud parts will not be amplified as much as the quiet parts.  If you compress first and nothing is initially close enough to clipping, everything will be amplified equally (which is the same as just applying some amplification).  By normalizing first, the loud parts are close enough to clipping that they will not be amplified as much as the quiet parts.  After compressing, the loud parts will actually be above 0 dB and you will need to re-normalize the recording so that when you turn it back to a fixed point representation (like 16 or 24 bit PCM), it will not be clipped.

[momule]

thanks for the clarification Guys..

Nick

[macdaddy]

the problem comes because different software programs call different processes "normalizing." (marc and i had a long discussion about this once)...

imo - changing the gain is fine, but normailzing and/or compression are not...

changing the gain for the whole recording is simply increasing the volume, so the loudest spot is close to 0db. this makes the loudest part as loud as possible without clipping, but leaves the dynamic range intact. replaygain does the same thing during playback (specifically album gain) if you do not want to alter your actual music files.

in my limited experience, if you have correctly adjusted your levels, it will not be necessary to change the gain...

normalization, "volume," etc. are all no-no's because they alter dynamic range by amplifying certain parts more than others. this isnt as big of a deal with acoustic tracks in the middle of the set - where it really screws stuff up is within a track, where there are quiet parts and loud parts.

[SparkE!]

Normalization does not alter the dynamic range of the .wav you are working on.  However, if applied track by track, it does alter the dynamic range of the overall show since some tracks will be amplified more than others.  If normalization is applied to each track separately, each track is amplified to the point where it barely hits 0 dB.  Since the amount of amplification required to do so varies from track to track, the quiet tracks are amplified more than the loud tracks and you have reduced the dynamic range of the show.  Hoever, if you normalize the whole show, all parts are amplified equally and the dynamic range remains intact.  This is equivalent to increasing the gain by exactly the amount that results in an audio file that barely hits 0 dB at least at one point.  The loud tracks are still the loudest and the quiet tracks are still the quietest.

[macdaddy]

This is equivalent to increasing the gain by exactly the amount that results in an audio file that barely hits 0 dB at least at one point.

in some software programs, yes it is. in some software programs it is NOT the same thing, but more like compression (smoothing out the dynamic range, and the globally boosting the signal).


my point is that you should be familiar with the EXACT processes you software is doing when processing the audio files..

[SparkE!]

What program does it like you're talking about?

[macdaddy]

What program does it like you're talking about?

if i remember correctly, wavelab does it like this, and so did the last version of sound forge i tried (version 5, maybe)

[Tim]

this thread is exhibit A for why I cringe everytime I see a .txt file that mentions that the taper performed some sort of DSP.... many tapers have no idea what they are doing to the files but they do it anyway

[SparkE!]

In wavelab, it only does what your are talking about if you are batch processing a bunch of files (who may all be individually amplified by varying amounts) and no version of sound forge that I've ever used does it the way you are talking about.  (I've used the normalize function in version 5.)

Really, the problem is when you normalize multiple tracks, they probably are not all amplified by the same amount.  If they were all part of the same recording, then the quiet tracks are amplified more than the loud tracks.

You definitely don't want to use the normalize function of a burner program because it will individually normalize the tracks.  For instance, Nero gives you the option to normalize during the burning process.  The only time that is useful is if the tracks are individual recordings.  If a single recording is tracked out, then you don't want to separately normalize the tracks.  You want to normalize the whole recording, then track it out.

[macdaddy]

this thread is exhibit A for why I cringe everytime I see a .txt file that mentions that the taper performed some sort of DSP.... many tapers have no idea what they are doing to the files but they do it anyway

agreed.

i dont mind adjusting the gain, but other than that, the resample/dither step, and tracking, i do no processing to my recordings...

[divamum]

To add some basic Q's to this very interesting thread (I'm just starting to try my hand at editing some of my stuff - acoustic, but with WIDE dynamic range - so this is really useful info!)

First convert it to a floating point representation with the highest numerical resolution possible (usually 32-bit floating point)

Can you explain this for complete beignners? What is "floating point representation",why do we need it and what does it do? Tx.

then normalize it, then compress it and re-normalize it.  You will want to normalize it first because you want the loud parts to already be close to clipping.  That way, when you compress it, the loud parts will not be amplified as much as the quiet parts.  If you compress first and nothing is initially close enough to clipping, everything will be amplified equally (which is the same as just applying some amplification).  By normalizing first, the loud parts are close enough to clipping that they will not be amplified as much as the quiet parts.  After compressing, the loud parts will actually be above 0 dB and you will need to re-normalize the recording so that when you turn it back to a fixed point representation (like 16 or 24 bit PCM), it will not be clipped.

Interseting and very helpful.  My question here is - where are the compression features in Soundforge (which is what i'm using right now - I have Wavelab as well, but haven't had a chance to try it yet)?  What sort of settings does one use? So far most of my editing has been done simply on a "that sounds better to my ears" basis, but I fear that I'm actually making a lot of elementary mistakes by having no understanding at all of the basic "technique" of how it works.  All comments helpful!  Since I'm opera, my biggest problem is that the dynamic range is very wide, and since nobody is monitoring levels while I sing there are usually compromises at one end or the other; while I don't like the sound that too much compression produces, a small amount will probably result in recordings which are closer to what we hear on commercial classical singing recordings, which might not be such a bad thing when I'm making demos etc...

Anyway, fascinating reading. Thanks for any further explanations!

[SparkE!]

Wav files are stored in a format known as linear PCM.  Essentially, they use a signed integer representation of the samples.  For instance, 16 bit audio has a maximum value of 32768 and a minimum value of -32767.  Floating point representations are stored not as simple integers, but as an integer and a mantissa.  Some of the bits are used to store the integer and the rest are used to store the mantissa.  You can scale floating point numbers over a larger range without a significant loss in dynamic range than you can integers.

Example:

An integer representation of one 16 bit audio sample might be 11643.  In a floating point representation, I could write that as:

1.1643 x 10^4

If I were to drop the level by 40 dB (that's division by 100), the integer would become 116, thereby losing the .43 fractional part of the number.  On the other hand, the floating point representation would not lose that fractional part:

1.1643 x 10^2

If you wanted to increase the gain by 40 dB, then the integer representation would clip at 32768  (you really want 1164300, but there are not enough bits to represent it in 16 bits), but floating point would be 1.1643 x 10^6.

So, you don't see a reduction in dynamic range associated with gain adjustment.

Now, I'm not being careful with exactly how floating point numbers are represtented by the bits used for each sample, but you get the idea...

For way more than you probably want to know about floating point numbers, here is a link that explains it better than I can:

http://stevehollasch.com/cgindex/coding/ieeefloat.html

(Thanks to Steve Hollasch for the information at this link!)

[leehookem]

My question here is - where are the compression features in Soundforge

wave hammer

[divamum]

BWAH!

So, are you saying.... there are none in SF, or that the available ones suck? Enlightenment for those of us of the n00b persuasion...  (I'm lol, though)

Actually, is there any appropriate place roun' these here parts to start a "mixing/editing tips and tricks for newbies" thread? I'd be really interested in learning more from you folks who know a thing or two .  (Or maybe there's some stuff elsewhere online to read that people can recommend?) I realise it's all highly subjective and osmething you just have to learn by doing, but I suspect it's more than simply opening the eq and hoping for the best....

[eman]

Two things- in case this thread isn't already dense and confusing enough. Here's my take on a couple of things I don't think were addressed.

One- using any kind of compression (not data compression like MP3 or FLAC but sound compression) will alter the sound of the recording. This is what FM stations use, and why when you look at an FM wav file it seems to be all the same level. I don't recommend doing this to a recording. If you do use it, consider saving an uncompressed version in case later on in life you decide you don't like compression. Compressors do have their use in live music production, and that is for the soundguy to add as he sees fit.

Two- If you have more than one set or file, normalize each set or file by the same amount, assuming you used the same record level for each file. Otherwise you may get the outcome that when you pop in the CD for set 2, or worse, when you hit the first track of set 2 on a CD, the volume will change signifigantly. It helps if your processing software tells you how much it will adjust the levels before doing it- check each file and figure out which one will have the LEAST boost, and use that amount for all of the files.

It is the soundguy's job to bring up the volume on the quieter songs and make any changes in volume between sets, IMHO, not the post processor's.

Here's a thought, too. While you can't add information by increasing the volume in post production, perhaps the playback algorithm (20 bit processing on a 16 bit wave, for example) will benefit from using larger numbers? It seems like it will be able to interpolate more precisely if there are more values available between samples. Or (less likely to me) will it remain the same or (least likely) get worse?