[music-dsp] Random Noise Reduction

Yaakov Stein yaakov_s at rad.com
Wed Jan 14 23:43:00 EST 2004


> Looking for pointers on random noise reduction which, given
> a good characterization of the spectrum, 

which spectrum - that of the noise or that of the signal ?

> will minimally impact the quality of the sound.  Looking for moderate
> reduction within the quality constraint and also to minimize
> the sound of noise pumping.
> 
> Trying to see what I can do to improve a noisy mic for low
> SPL recording.

If the noise is stationary and uncorrelated with the signal,
and you know both spectra, then your best bet is a Wiener filter.
This can be implemented in either time or frequency domains.

If it is uncorrelated but nonstationary,
and you believe that it develops in time according to some linear system, 
then Kalman filtering will help.

If the noise is relatively low and your signal is well represented by an AR model
(e.g. speech or harmonic music) then LPC prefiltering is the way to go.

If the noise is stationary and the signal is not always present,
then spectral subtraction might help (but destroys phase).

Jonathan (Y) Stein




More information about the music-dsp mailing list