As a professional audio guy, I'm very aware of the need to captureclean and accurate audio along with clean and accurate video. When on ashoot, I follow accepted practices by using a separate directionalmicrophone (or two), keeping the mic as close to the sound source aspossible, and keeping the record levels as high as possible withoutclipping. But even with those precautions, noises occasionally creepinto my recordings and must be reduced or eliminated.
I'm also asked to edit and compress other people's footage, andoften there's no telling what distracting noises or just plain badsound exists on their audio tracks. Recently, I embarked on a projectto clean up and transfer to CD several hours of live Cuban musicrecorded in the early 1950's on ¼-inch analog tape. When it¹sdone, I will encode the music for streaming on the owner¹swebsite.
All of these jobs call for some audio surgery.
Fortunately, there are several hardware and software tools that cangreatly improve bad audio tracks and deliver sound that is muchimproved over the original recording. The trick, though, is finding theright audio tool for the specific noise problem. To do that, you mustfirst break the audio problem into separate parts and then deal witheach individually.
Undesirable noises can be broken down into two categories: fixedfrequency noise such as that caused by AC hum, air conditioning units,cameras, or generators; and dynamic noise, which changes in spectralcontent and level over time. Classic examples of dynamic noise includecar and airplane pass-bys and wind and surf noise. Fixed frequencynoise is by far easier to repair than is dynamic noise, but both can beimproved.
You first need to determine if you're a victim of Bad RecordingTechniques. Poor microphone choice or bad mic placement can leave youwith an audio track that's muffled or indistinct. The use of acamera-mounted microphone can result in the same symptoms. Happily,this is the easiest problem to fix, perhaps requiring only thejudicious use of equalization and volume control to reduce the level ofnoise. Boosting frequencies around 2.5 KHz for men and 3.5 KHz forwomen with EQ will make the spoken word louder and will help it to cutthrough. Boosting frequencies around 6 KHz will increase sibilance (the"ssss" sounds in speech) and intelligibility, but it may also increasebackground noise and hiss, so a light touch is required.
By cutting frequencies below 250 Hz you may be able to turn up thetrack's overall volume without distortion. The telephone companydiscovered long ago that most of the audio information in human speechexists between 300 Hz and 3 KHz, so there's little harm in cuttingbelow 250 Hz for a talking head. But remember that EQ will affect allthe sounds in the recording; so a little bit goes a long way.
If the original recording level was low, then you may find thatboosting the volume during editing also boosts background noises. Thesenoises can include camera noise or handling noise, or environmentalnoises like air conditioning.
If the background noise is only audible in-between sentences, yourbest solution is to use a noise gate. A noise gate allows sounds at orabove a certain volume level, called the threshold, to pass through itunaffected. But when the volume drops below the threshold level, thenthe noise gate turns the sound off altogether. When the volume againexceeds the threshold, the noise gate opens up and allows sound to passagain.
A close cousin of the noise gate is the expander. Whereas a noisegate clamps the volume to a minimum level when it falls below thethreshold, an expander just reduces the volume to a user-specifiedlevel. It is a kinder, gentler noise gate that is used to increase thedynamic range of a track, and can significantly reduce backgroundnoise. Often a noise gate and expander are combined into one effect.Both can be quite effective at reducing or eliminating low-levelbackground noise.
If the noise has a fixed frequency, as with AC hum or airconditioning noise, you can often use filters to reduce or eliminateit. However, basic bass and treble filters are too broad for this work,and will audibly affect your program material. Look for parametricfilters that will let you control the width of frequencies affected(known as the filter's "Q"). Parametric filters can be set to cut thelevel of a very narrow range of frequencies, leaving the rest of yourtrack untouched.
Some software packages offer specialized filters for eliminatinghum. Apple's Final Cut Pro, for example, now offers a dedicated humfilter. In addition to specifying the frequency of the hum, you canalso choose to eliminate up to five harmonics of the hum. Gentleexperimentation is always best, and be prepared to hit the Undobutton.
Distortion and Clicks
Audio distortion is the worst of the problems, and the mostintractable. Distortion occurs when an audio input is too loud for theinput of the next stage to reproduce cleanly. Consequently the tops ofthe audio waveforms are clipped off, hence the term "clipping".Distortion generally imparts a "raspy" quality to the sound, and inextreme cases can make your sound unintelligible. While analogdistortion is the most common, digital devices are also susceptible todistortion. Analog devices distort in a gradual fashion as gain isincreased, but when digital devices distort they do so at oneparticular threshold level, and the result is instantunpleasantness.
The best way to eliminate distortion is to prevent it before ithappens. There are few remedies for distortion once the recording isdone, especially in extreme cases. If the amount of distortion isslight or the distorted segments are short, then you can use yoursoftware audio editor's Pencil Tool to literally draw smooth tops onthe waveforms.
But if the distortion is severe or prolonged and there's no chanceof re-recording the sound, you may be best served by taking yourdistorted audio files to a facility that has Sonic Solutions' NoNoise,which is part of their Sonic Studio HD producthttp://www.sonic.com/sshd_home.html.
NoNoise is a venerable computer program specifically designed toeliminate unwanted noises from audio recordings, and it is particularlyeffective when dealing with distortion. It can even generateresynthesized audio to replace damaged sound. NoNoise is also your bestchoice for dealing with dynamic noise that does not respond tofiltering. It is the industry standard in noise elimination, but aNoNoise system comes with a hefty price tag, so it's best to farm thiswork out. Depending on where you live, expect to pay anywhere from$60/hour to $150/hr for NoNoise work. Your budget may not like it butsometimes there is no other option.
Clicks and pops can result from bad audio cables, or channel breakupwith wireless microphones. Also, if you choose to use an old phonographrecording of music, you’ll likely find that it is full of clicksand pops. These can be handled in much the same way as distortion. Manysoftware audio editors include a function that will automatically findand repair clicks. I don¹t recommend allowing the program toautomatically remove all clicks, as this can destroy transients thatyou want to keep, like ³t² and ³p² sounds inspeech. You¹ll do better to find them one at a time manually.
Noise Removal Software
There are several specialized software tools designed to removevarious kinds of background noise. They can perform many of the samefunctions as NoNoise, but at a much lower price. For example, DART Profrom Digital Audio Restoration Technology http://www.dartpro.comis a program for Microsoft Windows that includes several filters, clickand hum reducers, and gain changers that can be chained together toprocess your audio tracks in one pass.
Ray Gun from Arboretum Software http://www.arboretum.com is aMacintosh-based program that takes a much simpler approach to reducingnoise, by providing presets for attenuating noise, removing clicks andpops, and reducing common hum. This approach may be easier fornon-audio professionals to use, but its effectiveness is somewhatlimited on anything but minor noise problems.
While these systems are convenient, care must be taken in their use.Post-production pro Mark Berger makes the point that, "piranhas arebetter than sharks.” This is just an aphoristic way of sayingthat many small bites are more effective than one large gulp. The ideais to use a little filtering, then use a little gain change, and thenuse a little compression. In this way each device is doing a littlebit, which it does best, and the sum adds up to a more pleasing,effective sound.
According to movie mixer Randy Thom, "One thing that most peopledon't know about the process of trying to get rid of noise, especiallyenvironmental noise (as opposed to electronically generated noise) isthat manual gain riding and EQ almost always play a part in what turnsout to be the most successful approach.”
The best technique is often to use a digital algorithm to reduce thenoise a few decibels. More than 6 or 8 dB will usually generateunwanted digital artifacts unless the noise is very narrow band. Andthen use some gain riding and dynamic EQ to reduce the noise stillfurther. Some choose to start with the manual part of the process, andthen apply the algorithm.
The best way to eliminate background noises is to make sure they arenot recorded in the first place. But if you find yourself having tosalvage an audio track, there are plenty of tools out there that canhelp.
Steve Cunningham is a voiceover actor and producer and principleof Acme Voiceworx.