I'm looking for an algorithm for hiss reduction of speech (and/or other aural) signals. Any ideas? Thanks, John
Speech Hiss Reduction
Started by ●April 14, 2004
Reply by ●April 14, 20042004-04-14
John Cristion wrote:> I'm looking for an algorithm for hiss reduction of speech (and/or other > aural) signals. > > Any ideas?Couple of things come to mind: 1) use a low-pass filter with variable cutoff frequency, and adjust the frequency as a function of the amount of high frequencies present in the signal (the input signal). 2) FFT -- eliminate the values that are below a certain threshold. You could make that threshold adjustable, or try tricks like detecting when the signal is absent and you just have background noise/hiss -- the mean value of the spectrum at that time can be used to set the threshold. HTH, Carlos --
Reply by ●April 15, 20042004-04-15
Carlos Moreno <moreno_at_mochima_dot_com@xx.xxx> wrote in message news:<> 2) FFT -- eliminate the values that are below a certain > threshold.I believe you wanted to say "above a certain threshold", right? --Frank
Reply by ●April 15, 20042004-04-15
Frank wrote:> Carlos Moreno <moreno_at_mochima_dot_com@xx.xxx> wrote in message news:< > >>2) FFT -- eliminate the values that are below a certain >>threshold. > > > I believe you wanted to say "above a certain threshold", right?No, *below* a certain threshold... Or do you want to eliminate the speech and leave only the background hiss? Don't know if the phrasing I used suggests something else, but the idea was: take the FFT; loop through the N values and replace those that are below a threshold with 0; then take inverse FFT (with proper overlap methods). Carlos --
Reply by ●April 15, 20042004-04-15
Frank wrote:> Carlos Moreno <moreno_at_mochima_dot_com@xx.xxx> wrote in message news:< > >>2) FFT -- eliminate the values that are below a certain >>threshold. > > > I believe you wanted to say "above a certain threshold", right? > > --FrankNo. Hiss has (we hope) low amplitude, and small components have small importance. By eliminating components too weak to matter much, the quality of signal might be improved. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●April 15, 20042004-04-15
"John Cristion" <John.Cristion@jhuapl.edu> wrote in message news:<c5k7at$aip$1@aplcore.jhuapl.edu>...> I'm looking for an algorithm for hiss reduction of speech (and/or other > aural) signals. > > Any ideas? > > Thanks, > > JohnJohn, If you look over the right interval, the speech will be changing characteristics while the hiss will not. In other words, the hiss is stationary, the speech is not. Exploit that. Maurice Givens
Reply by ●April 15, 20042004-04-15
John Cristion wrote:> I'm looking for an algorithm for hiss reduction of speech (and/or other > aural) signals. > > Any ideas? > > Thanks, > > John > >I have seen an adaptive filter design in the amateur radio literature (I think it was "QST", but it may be "QEX") that capitalizes on the fact that speech is fairly highly correlated while hiss is not. The adaptation is set to be fairly fast, and the filter passes the correlated part. This would have been about 10 years ago or so -- if you find it you may find the source material that they were working from. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com
Reply by ●April 15, 20042004-04-15
Carlos Moreno <moreno_at_mochima_dot_com@xx.xxx> wrote in message news:<Xvwfc.24947$VA1.99094@weber.videotron.net>...> Frank wrote: > > Carlos Moreno <moreno_at_mochima_dot_com@xx.xxx> wrote in message news:< > > > >>2) FFT -- eliminate the values that are below a certain > >>threshold. > > > > > > I believe you wanted to say "above a certain threshold", right? > > No, *below* a certain threshold... Or do you want to eliminate > the speech and leave only the background hiss?No I was thinking of cutting off the values at a certain threshold... like a peak limiter does, to suppress excessive peaks in the HF area. Now I see what you meant. --Frank
Reply by ●April 15, 20042004-04-15
"Carlos Moreno" <moreno_at_mochima_dot_com@xx.xxx> wrote in message news:KIifc.19024$uc7.212174@weber.videotron.net...> John Cristion wrote: > > I'm looking for an algorithm for hiss reduction of speech (and/or other > > aural) signals. > > > > Any ideas? > > Couple of things come to mind: > > 1) use a low-pass filter with variable cutoff frequency, > and adjust the frequency as a function of the amount of > high frequencies present in the signal (the input signal).This is a fairly simple algorithm to implement. For an example of a commercial product that does this, see http://www.behringer.com/SNR2000/index.cfm?lang=ENG. From reading the manual, you may be able to get a pretty good idea of how it works. As has been mentioned by others in this thread, more complex algorithms relying on statistical properties also exist. CoolEdit has implemented this and last I knew had a free demo available for evaluation.
Reply by ●April 15, 20042004-04-15
Frank wrote:> Carlos Moreno <moreno_at_mochima_dot_com@xx.xxx> wrote in message news:<Xvwfc.24947$VA1.99094@weber.videotron.net>... > >>Frank wrote: >> >>>Carlos Moreno <moreno_at_mochima_dot_com@xx.xxx> wrote in message news:< >>> >>>>2) FFT -- eliminate the values that are below a certain >>>>threshold. >>> >>> >>>I believe you wanted to say "above a certain threshold", right? >> >>No, *below* a certain threshold... Or do you want to eliminate >>the speech and leave only the background hiss? > > > No I was thinking of cutting off the values at a certain threshold... > like a peak limiter does, to suppress excessive peaks in the HF area. > > Now I see what you meant. > > --FrankIt's easier to describe in analog. It's called center clipping in time, and this is the equivalent in frequency, like H.H.Scott's Dynamic Noise Suppresser. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
