Tech Notes: Mastering

A friend of mine was asking me about mastering recently, specifically whether I master my tracks or not, so I explained my methods to him. It occurred to me it might make an interesting blog post.

What is mastering? Essentially it's the final step in preparing audio for the listener's ears. Generally it means taking a finished song and running it through EQ and compression so that it sounds big and beefy like you expect a song to sound on the radio. Now, I have a huge problem with how popular music is mastered these days, so the mastering I apply to my music is minimal at best. I make sure the EQ is how I want it at the mixing stage, and apply only enough compression to the final mixdown to ensure that the overall loudness of each track is roughly the same. No fancy squeezing of the bass, expanding of the mid-range, or sharpening of the high end that could give you papercuts.

When a multi-track mix is finished, the first thing I do is play it back with the VU meter running, so that I can see whether or not the output is clipping. I touched on clipping in another post, but essentially it means audio is going 'into the red' and being cut off (or 'clipped') from the output stereo audio file. If this is happening, the VU meter will tell me exactly how much audio is going above the cutoff point, and I can drag the faders down to compensate. The end result is a clean file with all the audio information present, but a much lower volume than intended. I refer to this as the 'pre-master.'

The pre-master is then loaded into my audio editing program of choice: Sound Forge XP. The first thing I'll do is open the Normalize dialogue and run a scan for two things: the maximum peak level of the audio, and the RMS (overall loudness). If I've done my job correctly, the peak level will be somewhere below 0db (maximum). The RMS will vary depending on how much I've had to lower the volume of the track. It's generally somewhere in the region of -20 to -15db.

Sound Forge: the normalize dialogue box. Note peak level and RMS.

This is the waveform representation of a pre-master from a forthcoming e.p. Note the peaks are not touching the edges of the box. This is good - it means it isn't clipping.

Waveform: pre-master.

The RMS isn't always a reliable way of determining how loud I want the finished track to sound. If a song has no drums, for example, the RMS will come out wildly different compared to one that has. From research and trial-and-error I've determined that an RMS of -15db is a comfortable level for my music. Sometimes this varies from track to track by a factor of up to 5db. Ultimately I rely on my ears for the final judgement.

In order to bump the volume back up to where it should be, without the audio clipping, a compressor is required. This is a bit of software (or hardware) that determines which parts of the audio are clipping when you turn it up, and smooths them out instead of cutting them off drastically like a pair of scissors. The result is much more pleasing to the ear, but care must still be taken not to use too much compression, otherwise your audio will end up sounding like a Ricky Martin record (gasp!), and the waveform will look like a solid blue line, instead of a nice centipede shape.

Sound Forge has several compressor plugins, and they can be puzzling to work with if you don't know the first thing about them. I rely on the tried and true Wave Hammer compressor. Here's a picture of the dialogue box and the preset I use, called 'Master for 16-bit,' which takes all the guesswork out of it.

Wave Hammer: Master for 16-bit.

Inasmuch as I understand it, the first slider is the threshold, or the part of the waveform the compressor is going to lift up by its boot-strings. The second slider is how much attenuation, or squashing, of the peaks is going to occcur. The third is for the overall volume you wish to apply to the track beyond initial compression. If, for example, I've had to pull the volume down severely on the pre-master, I'll experiment with the output gain until I get back at least the amount of volume I took away in the first place.

Waveform: mastered version.

Here is the mastered waveform. You can see it's considerably fatter than the original (by a factor of 5db to be exact), yet you can still see the gaps between peaks: the mark of a successful operation!

Once this is done and I'm happy with the final result, I normalize to 98% (this is just a personal preference, and brings any peaks hitting -0db down to -0.02), and apply fade-ins or -outs if needed. I might also add silence to the end of the track if I think it leads into the next one too soon. Oftentimes I'll have to make adjustments or master a track from scratch if I've got something wrong in the mix, so I've taken to making notes of just how much compression I apply and any other edits I make after the fact.

And there you have it: how a Manitou record is prepared for consumption without making anyone's ears bleed or speakers explode.

Field Recordings, Part 2

Here's part two of my article on field recordings - part one of which can be found in the archive. This will give you an idea how much work goes into preparing the sounds I've collected for use in audio projects.

Once I have the raw recordings transferred from the Zoom H1 to my hard drive, it's time to edit them. Because I record a lot of sounds these days, I've made it a habit to announce what I'm recording at the beginning of each file. Before I started doing this I had trouble keeping track - especially if I didn't get around to editing files right away.


Step 1: I load the raw file into Sound Forge, my audio-editor of choice. I then trim the ends a tiny bit to remove the click noises produced by the Zoom's record button. When recording, I take special care to make sure the audio doesn't 'peak' (a techy term for overloading the mic, resulting in a distorted recording). As a result, the raw recording is often very quiet. So I normalise (increase the volume of) the file to 98% (2% below 0db or maximum).

Step 2: I remove any rumble from the recording if it requires it. Some of the lower bass frequencies can muddy a recording, so as a rule of thumb I EQ out anything below 80hz, which is about as low as the human ear can hear. If there's traffic noise in the background or particularly troublesome bass frequencies I'll go as high as 225hz. Mild wind noise can sometimes be scrubbed out in this manner too.

Step 3: Dynamics/compression. Oftentimes with digital recording the sound you capture is but a shadow of what was heard at the time. Compression allows an increase in volume while preventing the high peaks (the parts of the sound that have hit the 98% volume ceiling) from distorting. It also lifts subtleties such as ringing sustains, reverb, etc... that are naturally present but too quiet to discern in the recording's raw state. The downside to this process is it makes the background noise louder as well. But luckily there's a secret weapon to deal with that in step 4. Compression works best for very loud sounds, such as metallic objects being struck. Softer sounds, such as birdsong and streams, etc... are usually left as-is.


Step 4: Noise reduction. Once I have my file optimised for volume, it's time to scrub out some of the annoying background noise. I use a program called Goldwave for this. Its noise reduction filter allows you to feed in a sample of the unwanted background noise and magically scrub those frequencies from the recording. It requires some restraint, though, because being heavy-handed with this feature can harm the recording rather than improve it. I listen carefully to each file and adjust the settings to reach a natural-sounding balance. In some cases I'll apply a preset called 'gentle hiss/rumble reduction' if noise reduction affects the sound too much. Unless the sound is too problematic to clean (in which case I fall back on some extra EQ or discard it) the end result is a nice clean-sounding file.

Step 5: The final step is to load the cleaned file back into Sound Forge for final editing. I cut out parts of the file that have microphone or wind noise, or other unwanted sounds that haven't been completely removed by the scrubbing process. This is also when I cut files containing multiple takes into single pieces or 'one-shots.' These can then be used as-is, or loaded into a sampler and played as an instrument.

I've worked with these techniques on and off over the years, but this is the first year I've combined them as an archival process. I'm learning things all the time about how certain sounds respond: how much compression to use here, and how much noise-reduction to use there. There's still no substitute for capturing the cleanest sound possible in the first place, but when you're at the mercy of a noise-polluted environment or a sudden gust of wind, a little software can do wonders.