Audio Processing Trick : Normalization, Noise Reduction,

Other Audio Processing Tricks
EQ and compression are the most common tools that audio engineers use to process their audio. But they're not the only tools, by any means. Here are a few more that you may find handy in your day-to-day podcast production.

Normalization
Normalization means turning the audio up until the peaks are at a given level. This can be very handy, for example, if your record level is a bit low and you want to turn it up but not so much that it distorts. The simplest way to do this is to use Normalization.

Figure 1 shows the Sound Forge Normalization window. It's very straightforward; you can choose what level you want the file normalized to using the slider at the left. In this illustration, you'll see that we're normalizing to -0.5dB, not 0dB. You can simply normalize to 0dB, but if you do, you have no headroom. If you're doing normalization as your very last step, you don't need headroom. If you're planning on adding EQ, you should probably normalize to -3dB or even -6dB to leave headroom for later EQing.


Figure 1: The Sound Forge Normalization window

Of course, this is assuming you're using Peak Level normalization, which analyzes the entire file and raises the volume until the very highest peak is at the required normalization value. If you have one small section of the file that is loud, and the rest is very quiet, normalization is going to get you only so far. You'll have to use some compression to even out the levels throughout the file.

Another kind of normalization that's very handy if you plan on playing music is RMS normalization, shown in Figure 2. RMS normalization measures the overall power of a file, as opposed to just looking at peaks. If you combine this with an equal loudness contour, RMS normalization gives all your audio files the same apparent level.


Figure 2: Use RMS normalization with an equal loudness contour to make all your music files sound as loud as each other.

Note : RMS stands for root mean square, which is a mathematical measure of the magnitude of a varying quantity. RMS is calculated by finding the square root of the mean of the squares of a set of values.

Audio with lots of low frequency content sounds louder to us, so even though two files may "look" the same in an audio editor (both have peaks at -3dB), one may sound much louder due to its tonal character. RMS normalization using an equal loudness contour takes into account how sensitive our ears are to particular frequencies and sets levels accordingly.

Gating to remove background noise
A noise gate mutes the audio signal when the level falls below a certain level. For example, if you're doing an interview in a room with a noisy ventilation system, you could try using a gate to get rid of the noise when you're not talking. Setting up a noise gate involves setting a threshold and attack and release times, similar to a compressor, as shown in Figure 3.


Figure 3: Using a noise gate to remove unwanted background noise

Gating is best used when the background noise is very low; otherwise, the gating effect will be noticeable, because oddly enough our ears will notice the noise disappearing, and this can be more annoying than a constant background noise. If you have a serious noise problem, you may be better off using noise reduction.

Noise Reduction
Noise reduction is a sophisticated process that attempts to remove noise from your audio without affecting the rest of the content. As you might imagine, this is a very difficult undertaking. Noise reduction algorithms usually work by taking a "noise profile" or "noise print," which is used to determine the frequency content and level of the noise. Taking a noise print involves identifying a section of the audio file that contains only noise. The algorithm can then use this information to analyze the file and remove what it thinks is noise. Figure 4 shows the noise reduction windows for Audacity and Sound Forge.


Figure 4: Noise reduction, using Audacity and Sound Forge

You can see that both involve a two-step process — first getting a noise print, and then running the resulting filter across the entire file. Noise reduction can be incredibly effective for steady background noise. However, be aware that the urge to remove background noise can sometimes lead to audio that sounds too thin. The key is to use it sparingly, so that your audio sounds cleaner but not antiseptic and hollowed out.

Advanced EQ techniques

EQ also can be used as a corrective measure. The techniques described in the preceding sections aimed to improve the overall sonic quality of your audio by focusing on what you wanted to be heard. You also can use EQ to remove extraneous noise that you don't want in the file.

Clearing up noise
If you recorded your audio in a noisy environment, you can use EQ to get rid of some of the worst noise. First, roll off all the low frequencies. You should do this regularly to all frequencies below 60Hz, because they're generally not reproduced by most systems. You can roll off more if you're dealing with serious noise. For example, traffic outside a busy city window will be audible as a steady low rumble, with the occasional siren or horn. If you roll off more of the bottom end, the sound of the traffic will be less audible.

Similarly, you can roll off high frequencies if you have noise problems such as air conditioning or tape hiss. In the preceding example, we boosted all frequencies above 5KHz to give the file some air; clearly, this is not something we'd want to do if we were in a noisy environment. Of course, we could have used the shelf to roll off all high frequencies above 8KHz or so and added a slight lift around 4–5KHz. Very little important voice information is in the range above 10KHz, so if you have noise problems, you can safely roll this off without any damage to the intelligibility of the audio.

Of course, in extreme situations, you can get pretty savage with EQ if necessary. For example, sometimes news reporters outside during a storm sound like they're talking through a telephone line. This is because the audio engineer at the station has rolled off all low frequencies and high frequencies, leaving just the mids.

Pops
Sometimes, a pop sneaks into your audio file, even if you're using a pop screen. If you zoom in to your audio file and look at the offending pop, you'll see that it's a very brief, very loud low frequency burst. You can highlight the offending word (or even syllable) and roll off the offending bass frequencies, as shown in Figure 1.


Figure 1: A pop is visible as a short, loud low-frequency burst, which can be fixed with EQ.

Experiment with different amounts of roll-off at different frequencies. You'll find that if you roll off too much, you get rid of the pop, but the result may sound unnatural. Find a good balance between removing as much of the offending pop as possible, while retaining the natural feel of the original.

Dealing with sibilants
Some people have a problem with sibilants, which are consonants like the letter "s" ("d" and "t" can also be a problem but usually nowhere near as much) that contain a burst of high-frequency information. This is audible as a whistling or, well, an "ess" sound. A little of this is natural, but too much is annoying. Some people actually whistle when they pronounce these consonants, which is an extreme version of the same problem. Because this is usually a very localized problem, it can be dealt with using EQ.

Sibilants usually are most prominent around 5KHz. Just like the preceding example where a word was highlighted and bass frequencies were rolled off to get rid of a pop, you can highlight a word and try to cut the offending frequency. Dealing with sibilants is more troublesome than popping, because our ears are very sensitive to high frequencies, and even if it's only a momentary change, our ears may notice that the sound has changed. Not only that, but if someone has trouble with sibilants, it usually manifests itself throughout the entire interview — and the letter "s" is very popular in the English language. In general, if you want to deal with sibilants, you should use a de-esser. That's a fancy term we audio engineers made up to describe a "frequency dependent, side chain controlled compressor." De-essers are explained in the compression section, which comes next.