normalize sample library (not mastering)

[ahlist]

I'm working on a sample library. The samples are of electronic instruments (synths, drum machines, etc.).

Would you agree that in this case it is a good idea to normalize all the samples (each hit by itself). I know that normalize will raise everything (noise too) but it seems to me that in the case of a sample library, the user of the library will appreciate all the files being at the same level.

My other option is to record through some sort of pre-amp, etc. to raise the levels before I record. I've avoided that out of a (probably misplaced) sense of capturing the "pure" sound of each instrument (no coloring from other devices).

Your thoughts?

[synthoid]

The library will be unusable if the multisample doesn't play back at the same perceived volume level independent of pitch. Imagine using a keyboard where each key had a different volume response to the same force of touch; it would be a nightmare to play smoothly.

So, you really don't have any choice but to normalize.

In spite of the nonsense you will hear about normalization, it is simply gain applied to audio, no different from turning the audio up or down with a fader. So there's no need to become distraught over it. When the user of the sample library makes the playback volume a function of key velocity, gain will again be applied to the samples, and that will also not be a tragedy. So don't worry: normalize your samples. Don't pin them all the way to 0dbFS though.

-synthoid

[ahlist]

Originally Posted by synthoid

Don't pin them all the way to 0dbFS though.

-synthoid

Thanks for your reply. What do you recommend, -10 or so? or is that too far?

Also, I am recording at 24 bit, 44k and will also batch convert 16 bit 44k versions. Would you recommend I do the bit conversion first and then normalize the 16 bit versions. That is what I am doing now and seems to make sense.

[synthoid]

Hi,

-10dbFS seems a bit too low to me. I say that only because most sample libraries are normalized to a higher level than that. If you have a good sample library, say something from Giga or the libraries that come with Logic (for EXS24), you might pull up some of their samples to see how they prepare them. I pulled up some of the EXS24 piano samples in Peak and they have peak levels around -6dbFS.

I recommended that you normalize individual samples so that the instrument has uniform volume across the keyboard; this is the right thing for drum and synthesizer samples, because the end user will use his own amplitude envelopes to control the volume. But I should say that if the instrument is naturally not this way, i.e., it is naturally louder in the bass than in the treble region, then you should rely on the recording process to get the levels right, and when you normalize, apply the same gain amount to all the samples (to keep their relative volume levels). It sounds as though you are not working with natural instruments however, so I don't think this applies in your case...

good luck,

-synthoid

[auxpot]

always, i said always normalized before you dither down to 16bit....normalizing (i.e. volume changes) are adding to the digital wordlength, so always do it while at the higher state of bits 24bits in your case)...

dithering down to 16bit should always be the last step in processing....

[peeder]

Originally Posted by synthoid

In spite of the nonsense you will hear about normalization, it is simply gain applied to audio, no different from turning the audio up or down with a fader.

False. When a file is normalized and saved back to disk it goes from internal resolution (32 or 64bit float or 48bit fixed) down to 24 or worse 16bit, and requires dither or some other conversion process. So a fader is different than normalization.

Normalizing to a peak value is truly worthless in all situations. Normalizing to an RMS value is nearly worthless, as RMS level and psychoacoustic loudness are only loosely related. Setting the levels by hand, ideally during the sampling process itself as a part of printing at the right level, is the best approach.

Your samples most likely are for different velocities/loudnesses and you will want the quieter samples to be naturally quieter. No sense in normalizing them up only to bring them back down.

[ahlist]

Thanks all for your replies. All have been very helpful.

I expand upon my first post by confirming that these samples are all very electronic and usually have no volume differences from hit to hit - drum hits, noises, and non-velocity synthesizers.

Again, thanks to all. Hopefully I'll get a good set done to post soon.

[synthoid]

Originally Posted by peeder

False. When a file is normalized and saved back to disk it goes from internal resolution (32 or 64bit float or 48bit fixed) down to 24 or worse 16bit, and requires dither or some other conversion process. So a fader is different than normalization.

Right, this is exactly the kind of nonsense I was talking about.

-synthoid

[sardi]

Originally Posted by synthoid

Right, this is exactly the kind of nonsense I was talking about.

-synthoid

How exactly is this nonsense?

[synthoid]

Originally Posted by peeder

When a file is normalized and saved back to disk it goes from internal resolution (32 or 64bit float or 48bit fixed) down to 24 or worse 16bit, and requires dither or some other conversion process.

Originally Posted by sardi

How exactly is this nonsense?

I take it back. It makes perfect sense as long as your D/A converters operate on 48 bit fixed point numbers and your hearing is sensitive to the roundoff error of the 25th bit. Peeder is a mastering engineer so this is probably the case for him.

The argument above, by the way, has nothing to do with normalization. It simply says that opening a file in, say, SoundForge, doing an operation and then saving it out again is destructive to the audio. If that's how you feel about your tools, I don't see how you can get any work done. But hey, that's just me.

-synthoid