How Broadband Noise Reduction Works
The Broadband Noise Reduction module uses a proprietary technique called Dynamic Audio Signal Modeling™ to intelligently subtract the noise from the digital audio file. Noise is removed with multiple downward expanders that linearly decrease the gain of a signal as its level falls.
The first step in performing broadband noise reduction is to create what is called a noise signature by selecting and analyzing an example of the noise within the source material. Using this noise signature, a noise contour line is created which is used to define the thresholds for the downward expanders that will perform the broadband noise reduction. The noise contour represents an editable division between the noise and non-noise audio signals.
At the same time, DINR also creates a model of what the non-noise audio signal looks like. DINR then attempts to pull apart these two models, separating the bad from the good—the noise from the desired audio. The noise portion can then be reduced or eliminated.
The noise reduction itself is achieved through the use of multiple downward expanders. The threshold of these expanders is set so that the noise signal will fall below them and be decreased while the desired audio signal will remain above them, untouched.
Once the signal level has fallen below the specified Contour Line (which represents BNR’s threshold), the downward expanders are activated and decrease the gain of the signal as its level falls. Over five hundred individual downward expanders are used linearly across the audio spectrum to reduce the effects of unwanted noise.
One of the psychoacoustic effects associated with broadband noise reduction is that listeners often perceive the loss of noise as a loss of high frequencies. This occurs because the noise in the higher frequency ranges fools the ear into thinking the original signal has a great deal of energy in that range. Consequently, when the noise is removed it feels as if there has been a loss of high-frequency signal. DINR’s High-Shelf EQ is useful for compensating for this effect. See High-Shelf EQ.
It is important to understand that there is a certain amount of trade-off inherent in any type of noise reduction system. Implementing noise reduction means that you have to choose the best balance between the following three things:
|
The amount of noise removed from the signal
|
|
|
The amount of signal removed from the signal
|
|
|
The number of artifacts added to the signal
|
DINR gives you a considerable amount of control over the above three elements, and lets you maximize noise reduction while minimizing signal loss and artifact generation. However, as powerful as it is, DINR does have limitations. In particular, there are two instances in which DINR may not yield significant results:
|
|
Cases in which the noise components of the audio are so prominent that they obscure the actual signal components of the audio.
|
|
|
Cases in which the noise amplitude of a 24-bit file is less than –96 dB. DINR is not designed to recognize noise that is lower than this level.
|