Overall song volume

The best thing to do is to use your ears and  to find a happy meduim volume so that all songs sound the same.
You will definitly need to use different compressor settigns for each song.

Use your ears IS the best way. Sometimes a song that shows visually on the meters is just plain louder than others, there's a lot of reasons for this but as said, each song will be different if it involved different instruments etc.
I use a handy tool that come with Wave lab called the Global analyzer which will show you the average RMS value of a selection. I'm sure there's a free plug in on KVR too if you look.  This is why having your songs professionally mastered is a good idea if you don't understand some of these things.  Like all things audio, you need to grow a few whiskers first and it all comes together. It's not something you can figure out in a few weeks.
Read , read ,read and yes , ask questions a lot.

+1 for calibrated monitors Child.

This, along with the K-System has resulted in a degree of consistency I've not been able to achieve before. Everyone should do it.

For a CD compilation, create a new project and import all the finished (mastered) songs into one track in sequence. Add a volume envelope to the track and insert a node right before each clip. Play through all the songs until you've identified the quietest-sounding song in the sequence. Use that one as your reference and bring the others down to match it by adjusting each one's automation node. If the quietest song is just way too low, go back to its project and raise the volume there.

Alternatively, you can export the original songs in an un-maximized state and do the volume-maximizing in the CD project. In that case, you'd automate your limiter threshold rather than using a volume envelope. Either way works well, but I prefer to master in place in the original project so that I can then do my MP3 encoding from the exported files.

Perceived volume is determined by average levels rather than peaks. The easiest way to measure that is Average RMS. I use the TT Dynamic Range Meter plugin to watch my average RMS as the song plays. This puts me in the ballpark. Occasionally I'll have to automate my limiter threshold if the average RMS spends too much time in the "squashed range", which I have personally defined as over -10db. I like my stuff to hover between -12db and -14db most of the time. Your own definition of what's acceptable will be a matter of taste and is very much genre-dependent.

Raising the average RMS is fairly simple, at least in theory. It's just a matter of pushing the levels higher into the master bus limiter. The limiter should be set to limit peaks to as close to 0db as is practical, which for us mortals is actually between -1 and -3db. Then raise the input level to the limiter until the average RMS is where you want it to be.

Where it gets un-simple is when you start to notice that the perceived volume increase comes at a price, which is reduced clarity and punch. After a small increase, everything starts to blend better. But as you continue to push it up that blend gradually turns to mush. Some genres are more forgiving than others, so exactly where that magical blend/mush threshold lies depends on what kind of sound you're after. You can get into trouble if the loudest song in your compilation is overcompressed - when you later bring it down to match the other tunes it will sound very weak. That's why it's important to somewhat standardize your average RMS from song to song.

Bear in mind that you cannot match levels in a CD compilation simply by looking at average RMS levels. Human sound perception is more complicated than that. You have to use your ears, preferably listening with eyes closed so you're not influenced by meters or waveform displays. But if each song's average RMS is within 2-3db of one another it will make matching volumes much easier.

One last thought: it really isn't necessary to get every song to match every other exactly. It's perfectly acceptable to have a slightly lower-volume ballad in between two higher-volume rockers. If you have something like SoundForge or Adobe Audition, check the average RMS values for each song in a commercial album that's been professionally mastered. Pick something in the same genre as what you're working on and see how much variance they've allowed. Depending on the genre, you might be surprised how much variation there is - or isn't.

Bit:
 
Thank you. Excellent and rare advice. But I must ask, when you write:
 
“Occasionally I'll have to automate my limiter threshold . .”
 
How do you automate a limiter threshold?
 
Does repeated increasing and decreasing of level in the digital domain have negative effects?
 
You also say that if your mix hits the master bus too low, you have to raise it.
 
“Raising the average RMS is fairly simple, at least in theory. It's just a matter of pushing the levels higher into the master bus limiter. The limiter should be set to limit peaks to as close to 0db as is practical, which for us mortals is actually between -1 and -3db. Then raise the input level to the limiter until the average RMS is where you want it to be.
 
What limiter do you use in the master bus?
 
Assuming you’ve set the relative levels of all your tracks to your liking, how do you push the levels higher into the master bus without having to raise the volume envelopes of each track by hand, which is—I know—tedious and very prone to error?
 
Do you put compression/limiting into individual tracks?
 
If my master bus level is too low, I sometimes raise the level of the bus. Is this a bad thing to do?
 
One last level question: I have an RME fireface 400 audio interface. When I play audio into it, I make sure it’s going in loud—often just below the red light level on the RME. In this situation, I see the levels of the armed Sonar Track meter bouncing near—and sometimes into—the Red.
 
But after recording, my Sonar clips appear at a really low level, and I have to normalize them. What am I doing wrong?
 
Thanks again for sharing your knowledge.

TT Dynamic Range Meter

bitflipper
For a CD compilation, create a new project and import all the finished (mastered) songs into one track in sequence. [..] Perceived volume is determined by average levels rather than peaks. The easiest way to measure that is Average RMS. [..] If you have something like SoundForge or Adobe Audition, check the average RMS values for each song in a commercial album that's been professionally mastered.
 
 
I used to do this in my audio editor, while Sonar has progressed in its mastering capability in the latest versions and additions and I am about to master a full album in a sonar X1 session. Is it possible to get the average RMS out of a stereo clip within Sonar or with suitable free utility?  Both TT and Span seem to be realtime plugins.
If not, I would go to the audio editor only for that, but that would be time consuming (sonar bouncing and exporting, opening editor, measuring and listening, then go back to sonar and adjust threshold in the limiter).
 
I understand that average RMS is not all, but it's a good starting point ,and really needed to do this job. I also use two levels for bang-bang-songs and for ballads (11/14 in 2010, might raise now) and I agree it's genre dependent.
 
When I increase level through limiter threshold I do not only check squash, dynamics and clarity, but I also check:
a) bass and drums presence over other instruments
b) voice level against other instruments
c) sibilances in high freq
so it's really useful to have chance to get back to the mix in these modern times.
Recently, for tracks that had to compete with others in loudness,  I found better results by applying small compressions on the way, from each track, to each bus and to the master bus, so that I could apply less amount in the limiter.
 
Thank you much for sharing your knowledge, your reply was of a platinum level, I hope someone helps me with the average RMS metering (I'm currently checking how to force sonar meters to get a very slow RMS..)




If you have something like SoundForge or Adobe Audition, check the average RMS values for each song in a commercial album that's been professionally mastered. Pick something in the same genre as what you're working on and see how much variance they've allowed. Depending on the genre, you might be surprised how much variation there is - or isn't.