Cleaning Your Noisy Audio

When we created our test setup for this project, we deliberately left our computers, routers, printers, and monitors running, producing a consistent din of ambient noise. Our thinking was pretty obvious: how can you possibly test a shotgun microphone's ability to ignore ambient sound without lots of ambient sound? Also, since few interviews or other types of videos are shot in sound booths, the noise also provided a lovely dose of reality.

Once we heard the background noise present in even our highest-quality recordings, our thoughts quickly turned to noise-reduction software. The software generally works best with regular, consistent background noise such as machine hum or tape rumble, as opposed to irregular noises such as cheering crowds or traffic noises. We surveyed our labs and checked recent press reviews, and rounded up three products.

First on our list was Sony's vaunted Noise Reduction filter for Sound Forge, which costs $280 direct, but requires Sound Forge ($399 direct) or another DirectX-compatible audio editor to run. Second was BIAS' newly released SoundSoap Pro for Windows ($599 retail) from BIAS, a DirectX plug-in we ran within Sound Forge. Third was the noise-reduction capability that comes with Adobe Audition, a complete audio editor that retails for $299, though it's much cheaper in several Adobe bundles.

Note that true noise-reduction filters work differently from inexpensive functions typically called noise gates. Briefly, the machine hum from my office equipment was most noticeable during the short periods between spoken words in the original recordings, though you can also hear it throughout the recording.

Noise gates work by setting thresholds for noise. When the audio exceeds the threshold, the noise gate filter assumes it's audio that is supposed to be heard and leaves it alone. When the volume goes under the threshold, it assumes the audio is noise, and simply eliminates it, reducing volume to zero.

When the speaker paused for breath in our test recording, a noise gate reduced the volume to zero, eliminating the hum during the otherwise silent period. However, noise gates do nothing to reduce the audible hum while the subject's speaking. This produces a noticeably artificial effect, since you can hear the hum when the subject's speaking, but not when she's quiet.

In contrast, all three noise-reduction technologies try to identify the background noise and eliminate it throughout the entire recording. You start by selecting an otherwise blank stretch in the audio file one that's devoid of speech, music, or other audio that's supposed to be there, and contains only the background hum. Then you tell the noise-reduction filter to capture a noise "print" or profile of that hum, and eliminate it from the entire audio file. Figures 2.11 to 2.13 show the noise profile captured by all three products.

Figure 2.11. Grabbing a noise print with Sony's Noise Reduction filter.

Figure 2.12. SoundSoap Pro's highly effective eye candy.

Figure 2.13. Adobe Audition's competent Noise Reduction filter.

We tested each noise-reduction technology on three audio files. We highlighted the same segment in the three original files to identify the noise print and then ran each filter using the default settings.

Two of the audio samples came from this project first, the low-quality audio recorded with the embedded microphone on the Sony camcorder, and second, the excellent audio captured with the Shure handheld microphone. We also tested an audio file recorded from a cassette player.

On the high-quality recording, all three technologies virtually eliminated the machine noise, making it sound like it was recorded in a sound studio. The Shure microphone produced a great original signal, and all three filters made it sound positively pristine.

The results with the lower-quality recording were less impressive; basically, you can't turn a sow's ear into a silk purse, no matter how good your software is. Though all three programs removed most of the background noise, the detail that wasn't captured by the camcorder's microphone was not magically restored. While the recording sounded better, the improvement was incremental, not exponential. The lesson is clear: get a good microphone setup and don't expect to fix it in post.

The results from the cassette tape were simply fantastic, with all three products dramatically reducing tape hiss and rumble. If you're doing this type of conversion, noise-reduction technology should be in your budget.

Which product performed best? Very tough call and my instincts are to be politically correct on this one and say they were all equal, especially given the focused nature of our tests. In addition, since we used the default setting on all three products, it's impossible to say whether a bit of fiddling with any of the products would have produced significantly different results. However, if you held a gun to my head (or threatened to crush my beloved three-chip DCR-VX2000) I'd have to say that SoundSoap was shockingly good.

Happily, you don't have to put your money down based on this extorted recommendation; all three vendors offer free, downloadable trial versions that you can run on your own noise problems.

Jan's Rules

Here are the rules I've learned to live by when working with audio equipment in the field:

• The first time you use a new piece of equipment you're going to screw it up. So, you should perform at least one test shoot with anything new before you actually use it in an important shoot.

• You can't capture good audio without a good set of headphones. Ear buds are tempting because they're portable and unobtrusive, but they let the outside noise in, making it tough to differentiate between what the microphone is picking up and what you're actually hearing. For best results, get big, clunky "noise-canceling" headphones that block outside sound.

• If you're working a concert or conference, bring a small flashlight and some kind of tape to secure your audio wires or power cables.

• If you're picking up audio from the sound system, test out the system a few days before the event to make sure you have the proper connectors. Then arrive early the day of the event, to hook up and test again.