normalization

klo wrote on 4/7/2005, 7:13 AM
can someone explain to what level should I normalize audio (0db; -12db or -20db peaks) or any other. taking that broadcast legal audio level is -20db peaks digital = 0db analog I assume that -20db peaks should be my level of normalization but in VV the max. is -12db.

My next question is - if I want to work on audio file like e.g. compress audio, clean, put the equalizer etc. should I do it before normalization or after.

Comments

Rednroll wrote on 4/7/2005, 10:59 AM
You have some misunderstandings going on. When they use a number like -20dBpeak=0dB VU, this is setup using a calibration tone, so that you can adjust the inputs to your analog gear to avoid distortion during a transfer. This usually consists of a -20dBpeak 1Khz sinewave being played back from the digital equipment, where you then adjust your inputs which are being fed from the digital sources outputs. As far as your audio levels are concerned, you should then be able to normalize to 0dBpeak and avoid distortion concerns. A VU meter in analog equipment responds to average dB levels. A peak meter used in digital, responds to instantaneous levels. There's a big difference there where -20dB=0dB is not a correct equation that you're using, thus you're leaving out the actual units in this equation. Each of those dB values has an additional unit associated with it. A dB is actually a unitless value until it has some other associated unit of reference tagged onto it. Thus it's actually -20 dB "PEAK"= 0 dB "VU", or 0 dB "SPL".

"My next question is - if I want to work on audio file like e.g. compress audio, clean, put the equalizer etc. should I do it before normalization or after."

Normalization is usually the last process, followed by "dither" if you use that. All the other processes you mentioned will alter the level of the audio, thus if you Normalize first, then you just undid the normalizing process, by altering an EQ, Compressor, cleaning the audio. So you would do these processes first then normalize if necessary at the end. Peak Normalizing works by scanning through the entire sound file and finds the highest peak within the audio. So let's say you choose to normalize to 0dBpeak. The normalization scans through the entire file and finds the highest peak value. Let's say it found it to be at -6dB peak. It now increases the level of the entire file up +6dB, so now that highest peak is at 0dBpeak. So you should be able to understand why you want to do this process last, because if you normalize first then add EQ and make a boast or cut, either you would exceed the 0dBpeak level by a boast, or reduce that peak level, so that you're no longer at 0dB peak. So you just undid the reason you choose to normalize.

Geoff_Wood wrote on 4/9/2005, 3:47 PM
After, othewise what is the point ? A good level to normalise to is -0.3dB.

All your talk of -20 etc relates to average power content, which is really little to do with peak-level normalisation, though RMS-power-weighted normalisation can be done (which includes compression if required).

geoff
scifly2 wrote on 4/10/2005, 1:06 PM
Normalize to just under 0 ( -0.1 ) unless you intend to submit a copy for broadcast, ( example - render to dv tape ) normalize that to -20.
rraud wrote on 4/10/2005, 5:33 PM
The major networks (at least here in NYC) request "digital audio tracks be referenced (1kHz sine wave) at -20dBfs with program peaks not exceeding -10dBfs"
Geoff_Wood wrote on 4/11/2005, 2:13 PM
If the normalise reference is -20dB, how can there be peaks at -10dB ? Unless there is 10dB frequency response ripple wrt 1KHz !!!

geoff
Rednroll wrote on 4/11/2005, 4:21 PM
From what I understood, was that the reference tone is at -20dBFS. The actual audio material would get normalized to -10dBFS.
rraud wrote on 4/12/2005, 9:29 AM
I was not really referring to normalizing, just trying to clarify the confusion over reference (test tone) level and program level, that has been a source of confusion (and poor audio) since digital audio first appeared... even before that....
As for normalizing for broadcast, Red would be absolutely corrrect.