In general, what's a good dB level at which to normalize my audio files?

Question

I'm a musician who extracts a very wide variety of audio samples from different sources. I then layer and arrange these samples together within music software to orchestrate songs.

When I save these extracted samples, I like them to be at roughly the same volume, so I normalize them. My audio editing software provides default options of normalizing to either -0.1 dB or -3.0 dB.

1. Why would I choose anything different than "-0.1 dB" – is there some danger of normalization that I should know about?
2. In general, what are some best practices to consider when normalizing audio?

Answer 1

In my experience, it all depends on where the audio is to be used next. If there is a limiter of some kind in the next phase of the process. Sometimes a limiter will be applied at varying levels below 0dB to prevent overload and clipping distortion.

-o.1dBFS is a peak normalization preset because basically, that is the highest level a sample can be without it being perceived as an overload in some hardware or software.
-3dBFS Is, I think a safe 'guess' as to where a limiter may be triggered in most common use.

I was told that limiters in nightclubs can begin to affect the signal at as low as -10dB. So if you were mastering for that use, you would set the peak at -10dB.

For storage and use in a library, I would say that consistency is most important. And at -0.1dBFS peak, you are saving the audio as close to full scale as possible.

Remember that peak level and perceived loudness differs. So your library of parts/samples each might have a different perceived loudness. But for inserting into projects, this is fine because level adjustments will be made in the project anyway.

Answer 2

If you normalize all of your samples to -0.1 dB, then use them in your DAW projects, you'll end up clipping your audio tracks all the time as soon as you use a plug-in.

I use Pro Tools where the channel faders default to unity gain in a new project. The channel fader on an audio track is post-plugin. This means if I put a Kick Drum sample on Channel 1, and it's peaking at -0.1dbFS, then I apply some channel strip plug-ins, it's quite likely it'll be clipping straight away. Now I have to go into the plug-in and adjust the input gain level.

I prefer recording at around -6dbFS, to give myself a little headroom when I use it. This leave me to bring up the gain on the mixing screen, where I want to be when adjusting levels.

If you're using something like Kontakt, to put all your samples in, you can just adjust the overall gain. However if you put individual samples through FX chains in Kontakt, you might get the same clipping problem. You'll end up turning down all the samples you normalized to the max.

This is completely personal preference though, and applies to your work flow method and the tools that you are using. There is no hard and fast rule. If -0.1dbFS works for you then it makes the most sense. It all depends on whether you find yourself turning every sample track up or down all the time, in which case they all need adjusting.

If I had to choose one option I'd go for the larger headroom of -3.0 dB, but personally I don't do this.

Answer 3

Best practise these days when considering content normalization is not to normalize to an arbitrary peak level, but to consider programme content and overall loudness when conducting normalization activities.

There are two areas of loudness that need to be considered:

1. Integrated loudness - Integrated loudness measurement (in LUFS) made according to ITU-1770-4 algorithm (gated loudness measurement that is integrated across the entire content)
2. True Peak level - True peak level considers the inter-sample peaks that can be generated by a DAC chip but are not considered by a discrete sample peak detector. (Achieved through up-sampling of the content to a higher sampling rate).

Loudness measurement standards are defined in EBU R128, ITU-1770-4, ITU-R BS.2054-4 and there are any number of available implementations of these standards and algorithms.

You should consider normalizing content according to the content standards of the platform where you will be delivering the material. For instance in Australia, Free TV Op.59 recommends that Branded TVC material be delivered at -24LUFS +/-1dB

Use of these loudness measurement techniques allows content to be easily altered to match surrounding content through simple gain changes.

You will find also that music content has different normalization conventions when considering loudness. It's also worth investigating how ReplayGain fits into this model in a music context.