Simply stated, you do not want noise to reach the post-processing stage. Do everything you can to avoid recording it in the first place, because once it's on tape, there will be no way to remove it without at least some signal degradation. In many cases, depending upon the type of noise, it cannot be removed at all. The following section, however, is for those situations when noise was unavoidable.
The budget method for reducing noise within a sample is to use a parametric equalizer to filter out the offending frequencies. Many waveform editors, such as SoundEdit 16 have a parametric EQ filter that can be used specifically for this process. To do it, you must determine at least roughly what frequency range the noise inhabits. Then, set the parametric EQ to ramp down that frequency range. Unfortunately, if the frequency range of the noise overlaps that of the desired signal, you're in trouble since you can't change one without changing the other. On the other hand, if you have a high signal-to-noise ratio, then the effects on the desired signal will be far less noticeable than the effects on the noise. In this type of situation, it is simply a matter of trial and error to see what is acceptable and what is not.
When using a sensitive condenser mic to record, one tends to pick up a high frequency hiss. This is one situation that can usually helped through EQ adjusts. Simply drawing down the high-end frequencies, one can greatly reduce the perceived noise in the sample. However, while this tends to work well for primarily low frequency recordings, such as male voices, it can have unwanted effects on higher frequencies such as those frequently inhabited by female voices. Also, this technique will often result in a decreased brightness of the recording, creating smoother, rounder sounds which you do not necessarily want.
Intelligent Noise Reduction
A much more expensive but far more effective method for noise reduction is to use software such as the Digidesign Intelligent Noise Reduction (DINR) plug-in. This software module works in conjunction with Digidesign's Pro Tools software and other compatible programs. It provides removal mechanisms for two specific types of noise: hum and broadband.
Hum is noise that can be classified by just one or two frequencies plus its harmonic and subharmonic frequencies, such as that shown in the image below. This kind of noise is often caused by things such as air conditioners and fans. The DINR plug-in allows you to specify a specific frequency to be removed. It then analyzes the sample to identify harmonics and subharmonics of that frequency, targeting them removal automatically. From this information, DINR sets up a collection of spot filters each of which is able to remove a very narrow frequency band. The user can then listen preview the changes in real-time, comparing the modified sample to the original. In this way, the settings can be refined for maximum effectiveness.
As the name implies, broadband noise is noise that covers a broad range of frequencies. Endemic noise such as tape hiss tends to fall within this category. DINR allows the user to identify a section of the sample that contains nothing but the noise itself. From this, the software analyzes the noise and creates a frequency distribution of the noise, as shown in the image below. This distribution serves as a continuous signature of the noise, which DINR can then systematically subtract from the sample as a whole, removing the sonic energy of the noise source. Again, the user can preview the changes in real-time, comparing before and after versions of the sound. After refining the settings for maximum effect, the sample can then be processed to remove the noise.
Unfortunately, even high-end noise reduction systems like DINR have their limitations. Most systems are limited to working with fairly continuous static noise such as hum and hiss. There is very little that this type of system can do for very dynamic noise, such as voices talking or other noise sources that contain a wide range of changing frequencies. The lesson, of course, is to try not to record the noise in the first place because once it is there, you will have a tough time getting it out.
For more information on the DINR plug-in, see: http://www.digidesign.com/Newdigiweb/Digiprod/dinr/moreinfo.html
Last modified 4/26/97.