9.51 - Impulsive noise
Description
This article is from the Audio Professional FAQ, by with numerous contributions by Gabe M. Wiener others.
9.51 - Impulsive noise
In this category we include many types of disturbance, from the click
generated by a scratch on a 78rpm disc, to the tiny tick created by a
single corrupt bit in a digital data stream. Also included is crackly
surface noise from 78's (that sounds like a frying pan), though this
requires somewhat different treatment; however the outline presented
below is fairly similar for both processes. Typically, audible clicks
are of a few microseconds to a few milliseconds in duration, and their
density can be up to a few thousand clicks per second on poor-quality
material.
First the audio is split into short blocks of maybe 10ms duration. A
model is fitted to each block; this model can be thought of as a
description of the signal in simple mathematical terms. The model is
chosen such that musical data is a good fit, but the impulsive noise
is a poor fit.
For example, a simple model could be a sum of sinewaves, whose number,
frequencies, amplitudes and phases are the model parameters. The para-
meters are calculated such that when these sinewaves are added together
they match the musical parts of the signal accurately, but match the
impulsive noise badly.
Now the model can be thought of as a prediction of the music. In
undamaged sections the prediction is close (since music is known to
consist of a sum of sinewaves, at least approximately); during clicks
and pops etc. the prediction is poor, because the model has been
designed to match the music, and not the noise.
Now we can achieve impulsive noise removal by replacing the data that
fits the model badly (ie the clicks) with data predicted by the model,
which is known to be a close approximation to the music.
Continue to:
My Books
