Adjusting Volume Levels of Audio Tracks

By Kurt Zawiedek

Mastering a Christmas Compilation, I had the task of combining tracks of very different provenience. Old recordings (Thornhill of 1937) met brandnew tunes (Diana Krall of 2010). A dadaistic choir recording of Christian Steyer juxtaposed Frank Sinatra crooning.

The selection of tracks was made by browsing youtube for two evenings, but as youtube blocked several of the most interesting tracks, some final decisions had to be made with the 30 second previews of the mp3 store.

After a basic track list was compiled following the moods from Christmas Eve to New Year’s Eve, I was puzzled how to get those differing tracks onto an adequate level for decent listening. Which ones had to be lowered in volume, which raised? Would EQ adjustments have to be made?

On previous projects such as Broken Windows I used the capability of the metaflac library to calculate replay gain and sort out the appropriate volume adjustments with the help of a spreadsheet. However, this approach would have involved the conversion of all mp3’s and wav’s into flac files. There had to be an easier way.

And indeed, there was. Step 1 was SoX, the “swiss army knife of sound editing”. The command line tool comes with a stats effect that calculates RMS levels, among others. The numbers showed that -20dBFS was a good RMS target level. The tracks were mostly jazz recordings, thus not maximized to the extent rock-pop productions are.

Step 2 was the command line tool “normalize-audio”. It automatically calculates ajustment levels for a set of files based on RMS and applies those gain adjustments as replay gain tags for mp3 or directly on a wav, including limiting.

This approach proved very helpful, as almost all tracks were smoothly leveled to a continuous volume. This simplified the task of finding those spots where manual adjustment became neccessary. The were three of them: One track had to be leveled down one or two ticks because the programme was an instrumental solo – thus, it had to be less loud as listeners expect a small group of unamplified instruments to sound quieter than a big band.

Two consecutive tracks still seemed to differ in loudness: Guaraldi’s trio sounded louder than Glenn Millers orchestra. A closer look revealed that Guaraldi had an extraordinary heavy e-bass in his combo whereas Miller had a flat sounding dog house double bass. The aural difference was eased by lowering the bass on the first track and increasing the bass and low mids on the second track.

Both sox and normalize-audio are great tools. In the future, I might use sox more often because of its large feature base and extremely quick execution. The other one, normalize-audio, is very easy to use even for command line beginners. It’s always nice to have your mix CD mastered to equal loudness.