Continued....
The quiet piece will have peaks around the same level as a loud piece??! It can be tricky to get your head around, but this one little concept is really quite important to understand. (especially when we start to think about sample libraries that we use when composing - I'll get to this later!)
Both these recordings ALSO go through another mix (and master) stage before being played on Spotify or Radio or where ever. This is because people like to listen to music in a way that they are not turning it up and down constantly, even if the piece is quiet, or super loud. People like to hear things at approximately the same level all the time.
For radio / spotify etc, this average level (compared to our 0dBFS maximum) is quite loud. So a mix engineer and then mastering engineer will do a tonne of things to make sure that the average listening level is at the intended level for the audience. It is a REALLY important part of the process, and the part that you are coming up against with your question. (EQ can also be dealt with by mix AND mastering engineers.)
So here's our situation for spotify/itunes/radio etc. The long term (approximate) average level of a mix needs to be set at a particular point so that the perceived loudness is at around the same level as similar types of music. (Hiphop will have a louder average than classical, but I'm trying not to get too complicated here!)
There are many many ways to measure average level. In the past, VU's were used (and were extremely useful for analog audio and for many folk still useful in digital audio!) . We have newer tools dedicated to digital audio that also can help like a loudness standard known as LUFS which I personally think a lot more folk should use / know about!
There's even MORE reasons to start to try and understand all this when we start to think of different delivery formats. If your music is going on an ad, what level do you deliver it to the mix engineer at? There's more than one right answer to that - depending on a number of factors. But arguments around it come down to the fact that ads are delivered to different mediums at very different levels.
For TV in Europe, the final ad needs to measure -23dB LUFS. In australia, it needs to be -24dB LUFS (and they are all quite strict about this!) . Radio can be -1 dBFS, Spotify is now asking for -14dB LUFS. Cinema needs another different level entirely (measured sometimes by a dolby method, other times just by what the ad distributer wants!).
For music deliveries, different folk want different amounts too. If I deliver a cue to a mix stage with 0dBFS peaks and loudness around -12dBLUFS, the rerecording engineer will probably pull her hair out. (Thats VERY loud, and will have a very small dynamic range...) yet that is sometimes what is delivered to play on radio.
I realise I've spoken a lot about different ways of measuring loudness without explaining them. It is worth looking up and understanding the difference between digital peak metering (dBFS), analog average metering (VU), digital loudness metering (dB LUFS) and other forms of metering as well (BBC meters, K metering etc!). Waves has a nice primer - there are probably many better ones out there. It was just one of the first google searches that came up!
https://www.waves.com/loudness-metering-explained
I guess the takeaway from all this is : When you listen to spotify, or just a file on a phone or computer, there is a maximum PEAK level that can be played back. (0dBFS)
The closer the AVERAGE level is to this amount, the louder the music will be perceived to be.
ASIDE : Adjusting volume on a computer / phone / device.
If you have a piece of music playing back with peaks of 0dBFS on a phone, yet it is playing back softly, it means somewhere in the system, there is a gain change going on. This gain change can occur either digitally or in the analog domain - but these days its normally digitally.
What happens is this. Your device has a maximum digital level. Thats when you turn the phone up to maximum. When this happens, your 0dBFS peaks will be played back at 0dBFS... and be converted to analog from that level (headphones are analog devices!)
When you turn it down, its still reading the audio file peaks at 0dBFS, but the level is being reduced by a simple math algorithm - say to -20dBFS or even much softer depending on where the volume control is at.
So say you deliver an audio file with peaks at -10dBFS, and your volume on the phone is set to -20dBFS. (You don't get to see that number - it might just be around half volume!) Your audio file will sound softer than one delivered with peaks at 0dBFS (assuming that its the same piece of music or just has similar overall dynamic range!). Indeed it will sound 10dB softer - which is quite a lot!
I did mention earlier that I would talk about levels in sampling.
Its quite interesting. I spoke about recording engineers setting equipment levels to record for an instrument - and even the song that is played on that instrument. With sample libraries, the sampling engineer doesn't know how the samples will be used. So she needs to record the samples so that the softest and loudest samples all fit within the required dynamic range. Some sample libraries compress the dynamic range too - but many don't too much or at all! This means that for a piano, the softest notes will play back at a very low level in your DAW indeed. I just tested one here - and the softest note of a soft patch was below -50dBFS peak. The loudest was -3dBFS. Thats a BIG dynamic range. And is from memory around the range of a concert piano. I'm lucky enough to have a really lovely piano to record, and I've only ever had around 35dB range from my somewhat uncooth playing! I'm not that controlled! Ha!)
So then, play that back in Kontakt. Often library developers put the kontakt fader at -6dBFS by default. Why? So that if you play back a number of sounds together, they don't peak straight away. (When you add sound together, the peak levels can increase. Loudness increases by 6dB (measured) when two identical, coherent signals are added together). So they're being slightly - but not TOO - conservative with levels, trying to make it easier for a user to not cause peaks.
What this all means is playing back some super soft velocities on piano can mean very low average levels compared to digital zero (thats 0dBFS!).
And this is where your mix / mastering decisions come into play - that others in this thread have started to talk about.
Normalising is just an algorithm that looks for your biggest peak (lets say its -12dBFS) and brings it up to another value (say 0dBFS). So then it applies a gain of 12dB to the entire track (equally to EVERY sample) and the overall volume of your track is increased. There is nothing wrong with doing this if needed. Its a great tool. Its just gain. You NEED to change gain for different delivery formats.
But you can do more than this. You can reduce slightly dynamic range using compressors. This can be done almost imperceptibly, or horribly (or for deliberate artistic effect) depending on how you use the compressor and exactly what the compressor is doing (they don't JUST compress a lot of the time... and there's tonnes of different models for compressing!) . You can equalise to help balance the frequencies you hear (which WILL change your dynamic range slightly) - or allow slightly more compression at times - or so many different creative things.
Generally speaking, you balance for tone, then compress slightly, then bring the level as close to digital zero as possible, and then measure for loudness (say in LUFS!) . Is the measured LUFS level what you want it to be? Great. If not, what decisions can you make to change it. Note : you can also automate the level of your whole performance - which is like a really SLOW compressor - in order to reduce dynamic range if needed. Once you get into mixing, theres a BUNCH of things you can do / try.
So there you have it. I've skipped over many things, probably oversimplified some bits, over explained others. But hopefully it gives you a starting point to learn some of the concepts. And to ask more questions. Never stop asking. My old assistant - now our head mix engineer here, went from knowing not much about digital audio (even with a masters in film sound design) to going back to university this year to study acoustics part time. His questions keep coming - but damn he's learning a tonne and doing some awesome work!
Cheers!
*I've made LOTS of assumptions and shortcuts all thru this post. This is a fun one though, since it is possible to measure above 0dB - intersample peaks - the calculated level of audio between samples - can go above 0dB, though we try avoid this as much as possible!)