Video by WaveArtsAudio via YouTube
Source
The first step is to identify a section of your track that contains only background noise. Often this can be found at the very start of the track before the signal comes in. However, if the track has already been cropped to eliminate leading and trailing signal, then you will have to find noise during a silent section in the track. Dialog can learn a noise profile with as little as 50 msec of noise, so this could be found between words in a dialog track for example.
The learn time is specified using the LEARN TIME parameter. It has various choices from 50 msec to 5 seconds, and also has a manual mode. Choose the longest learn time that is shorter than the portion of noise you have identified. In manual mode, learning begins when you click the LEARN button and ends when the LEARN button is clicked again. All the other modes learn for the specified amount of time after the LEARN button is pressed.