'Normalise' tracks

There are several problems with matching loudness this way. It's not actually what we refer to as 'normalizing'; that's a different process involving nothing more than setting the max peak value of a file to a predetermined level. What follows is my own experience of this process, and is certainly not endorsed by Adobe (who were a part of the discussions that took place regarding it)'

For a start, the way that these matching processes work is that they are roughly based on an Leq method of establishing the levels in a file, with some frequency weightings applied, and at best this only gives a vague impression of how loud anything actually sounds - and it's relatively easy to fool the algorithm - as your customers have found! Originally it was introduced as a method of leveling up the sound in TV programmes, which also involve speech - but as a lot of broadcasters demonstrate regularly, it doesn't work very well for that either; I still have to crank the levels up and down between programmes, and that's exactly what it's supposed to prevent the need for. The other primary motivation, and this is only slightly more successful because of the extremes involved, was to prevent adverts sounding way louder than programme material.

What the algorithms take no proper account of is the content, and its frequency and level distribution of a music file. Also, the measurement doesn't really take account of production level changes within it - you just get an overall LUFS figure. With the best will in the world, any songs you put on a track with an actual dynamic range are still going to end up with that dynamic range - and if you put them next to a song where it's been maxed out, then the quiet bits will sound quiet.

There's a complete 'ministry of truth' about all of this - a typical example (which completely fails to mention the major shortcomings) is here:http://productionadvice.co.uk/lufs-dbfs-rms/

The reality is that if you want everything to come out at similar levels, then you have to do one thing in particular - and that's to reduce the dynamic range of everything you are including so that essentially, there are no quieter bits. You can do this with a multiband compressor. It sucks. And at that point, it's easiest to level everything else up by ear.

Now, people might try and tell you all sorts of other things about this, but fundamentally, using an automated process to level songs out really never works properly, I'm afraid. It's generally easier and about as quick to do it by ear - but if songs contain quiet bits, you have to process them out to avoid complaints.