Am I creating even more distortion by clipping the signal before its normalized?
What does chopping or flattening the peaks do to the sound?
Is this better or worse than compression?
Worse!!! Clipping is distortion. If you get enough distortion, you’ll hear it. If you crank-up a 10 Watt amp and try to get 20W out of it, that’s clipping. If you want to know what digital clipping does to the sound, use Amplify and boost the peaks to +12dB or more. Then, save that as a “regular” 16-bit WAV file and play it back. (Audacity won’t actually clip internally because it uses floating-point, but WAV files will clip, or you can clip your digital-to-analog converter by playing a floating-point file that goes over 0dB at maximum volume.)
Clipping is a kind of dynamic compression… It reduces (compresses) the dynamic range by boosting the overall volume without boosting the peaks. But, it creates more audible distortion than “regular” compression.
It’s a matter of taste, but compression damages the dynamics! Music is supposed to have loud parts and quiet parts. By using compression, you can make it constantly-loud. IMO, most modern music is over compressed… This can make an “exciting” 1st impression, but it quickly gets boring and soon you are reaching for the volume control to turn it down. Compression in modern recordings is one of the reasons some people think vinyl sounds better than digital. (Not me… I can’t stand the “snap”, “crackle”, and “pop” of vinyl records!)
I create compilations with very different kinds of music from a wide variety of sources. To match the volume between selections, I use normalization, extensive use of envelope tool, compression, and compression based on peaks.
The basic problem everybody has is this: Many quiet-sounding songs have 0dB peaks. If you boost the volume (linearly without compression), you’ll get clipping. Since you cannot boost the quiet songs (or you can’t boost them enough) without clipping, if you want to match volumes (without compression), you have to reduce the volume of loud-sounding songs. Then, if you want to listen louder, you can turn-up the volume during playback time.
As you’ve found out, normalizing won’t work… Normalizing works on the peaks. Perceived loudness is more related to the average level and the frequency content. It’s best to match volume by ear.
Here’s the basic procedure when making a compilation:
1. Normaize all songs individually for 0dB peaks.
2. After normalizing, choose the quietest-sounding song as your reference.
3. Adjust-down the louder-sounding songs (by ear) as necessary to match the reference song.
There are tools such as [u]ReplayGain[/u] (and MP3gain and WAVgain) that try to automatically match perceived loudness. Of course they also tend to reduce the volume of most songs and on a compilation you could end-up with your quietest-sounding song not hitting 0dB. (For playback, I use ReplayGain with Winamp and it works great!)