The short version:
Is there a way to balance all the samples in a Kontakt instrument to (relatively) the same perceived (RMS?) level?

The long version:
So when I switched to Reaper from Logic I discovered that EXS (.exs) Instruments could be converted into Kontakt Instruments, which was great because I had, over the years, made and depended on a bunch of them.

But then I got on a roll and ended up creating my "dream drum machine" Kontakt Multi Instrument. It took a week of (going through every sample library I had acquired over the years and) selecting my faves samples, and then arranging them into velocity layered Instruments. The "Kicks" instrument alone has over 900 samples in it. The problem I'm having is that these samples vary hugely in their peak and perceived levels. When I'm trying to write a groove with them the "louder" samples naturally cause me to lower that channel's level, which then leaves the reasonably leveled samples too low and thus seemingly uninspiring. I am aware of this phenomena, but it's still easier to appreciate the sample that pops more easily than the sample that can barely be heard.

So, as stated at the top, I'd like to be able to balance all these samples so their perceived (RMS?) levels are approximately the same. As there are over 3000 samples in the entire Multi Instrument I'm hoping that there's an "automated" way to accomplish this - at least per Instrument (there are 12 Instruments in the Multi).

 

Last edited: May 7, 2017

Would be easier to take the samples out of kontakt 1st, then normalize the wavs with a RMS-aware tool, like the replaygain thing from foobar2000

 

Batch normalize those samples in SoundForge. Fully functional 30 day trial from Magix, "try before buy" version in sister site.
What i don't understand is how these sample libraries were stupid and yet they contained your favorite samples. Unless of course you like your samples stupid lol.

 

Thanks for the SoundForge tip - I'll check it out. Hopefully it can batch normalize to some kind of RMS level rather than just peak levels - I'll find out!

Of course the libraries aren't stupid. That was simply a poorly executed turn of phrase. What I meant to infer as being stupid is my cumpulsive collecting behavior (of the insane number of these libraries I've acquired over the years) . . . That said, the blistering level at which a few of these libraries package (their otherwise very interesting) samples is also kinda stupid.

 

Thanks - I'll check it out!

 

Got it done by taking the 958 samples outside of Kontakt and batch converting them with Twisted Wave. The results are of course imperfect as RMS isn't great with short, transient-heavy samples, but it has definitely brought them all closer than they were before.

 

Usually, RMS normalizing applies compression to the samples therefore changes the sonic character of the audio material that is applied on. I was only talking about peak normalizing, but since you are satisfied with the outcome, well done heheh

 

Interesting. I was not aware of this.

 

I'm not aware that it does, and most shouldn't, that's the definition of normalization, unless the 'compression' is clipping/limiting due to turning up & hitting 0dB when saving to fixed point.

 

RMS normalization makes an assumption of the softest and loudest passages difference in the sample, and then tries to bridge the gap. While this should be more correct than peak normalizing, because it is closer to how human ears perceive loudness, more or less it is like compressing the signal, when lower passages get louder and the loudest passages stay unaffected. The big difference here is, as humans we perceive different frequencies of the same volume, as louder or quieter but they aren't really. Peak normalization will just raise the volume of the signal according to the highest peak, so it will produce the loudest signal possible without affecting the dynamics within the signal, but again if you peak normalize 5 samples to 0db you will practically be out of headroom when these samples play together, your meters in your daw will go red hehe. So it's always what method suits you better with the audio material at hand, and to be true , imho, none is perfect

 

I've had a closer look at my samples pre and post the RMS normalization and there has been no compression applied in any way. Nothing has been changed except the level of the samples. The absolute hottest samples are now peaking at -3, which suits me fine.

 

Excellent. Btw, Junh1024 was right too. I am happy to hear the algorithm worked as it should because of course we 're talking about one shot samples, right ? My -somehow mistaken quote- was referring to recorded material like a loop containing more than one instrument etc that may alter its sonic signature due to very loud rms normalizing. Sometimes i get lost in translation and thanks to both of you for keeping the heads up with me. Cheers
PS: an interesting read
http://www.learndigitalaudio.com/normalize-audio

 

Thanks for the link.

Like I said, RMS normalizing and one shot, single-transient samples certainly are not a match maid in heaven. Much of how the RMS detection chooses the normalization level per sample becomes dependent on the length of tail after the transient. The longer the tail the lower the RMS value is detected to be, and so the hotter the sample is outputted through the normalization process. I experimented with a bunch of different settings and in the end chose to normalize all to -18 RMS with true peaks no higher than -3. The outcome is not perfect but better than what I had.

 

Ok then. THen it should follow that if you want a evener loudness, then you should find a workflow for truncating/padding every sample to the same length (say 5s), right? Unfortnuately, I don't know of any tools to do this.