On (02 Jul 95) Ron Newman wrote to All...
 RN> i've looked in my electrical engineering textbook at several
 RN> methods, but they all require a periodic signal, which audio
 RN> isn't since it's constantly changing, millisecond by
 RN> millisecond.
Here's my understanding of the way the NoNoise system works:
First, the audio signal is broken into small chunks of time, each
perhaps a few seconds long.
Each chunk is separately processed.  An FFT or similar transform is
used to convert the time domain sample to the frequency domain.  The
output of this step is a frequency spectrum.  Then each individual
frequency band is compared to a threshold level.  If the energy
level at that frequency is less than the threshold, that frequency
band is reset to zero.  This is exactly the same as using a high-Q
filter to isolate one particular frequency band, then running the
selected frequency signal through a noise gate.  The processed
frequency domain signal is then put through a reverse transform to
recreate a time domain sample, which is concatenated with the
previous chunk's processing to form the output waveform.
If there is a university library in your area, look in the music,
electrical engineering, math, and computer sections for information
on DSP algorithms such as this one.
--- PPoint 1.92
---------------