I have cetain sampled sequence out of non-stationary audio signal featured
by certain noise pattern (which is also non-stationary in details but is
well defined in general). I intend to run numerous recordings featuring
this noise, to separate the noise pieces ad process them in order to
obtain some kind of averaged noise stamp (in time domain) and then to use
it by cross-correlation with an actual signal in the buffer (real-time
sampling and buffering) in order to identify this noise pattern to be
filtered out by means of subtraction.
The noise pattern can be distinguished quite clearly but isn't exactly the
same in terms of samples in the tests, so that I thought to collect an
amount of this noise instances, average out them in sample-by-sample
manner and then use the result as the noise "stamp" to run the
cross-correlation with future signal in order to identify noise instance
and filter it out bu subtraction.
Do you think producing the noise "stamp" by averaging out the noise
instances in recorded signal is an effective way to follow in my case ?
I realize that the cross-correlation will never produce a perfect match
due to the noise "stamp" being a statistical (averaged from many
recording), but perhaps the maximum will reflect the best match. Am I
Thanks in advance, Alex