Noise Reduction Introducing Artefacts

[MikeS]

Using spectral subtraction on recordings to remove hiss, hiss removal is effective. However additional sounds seem to be added. The sound a bit like crinkling of silver paper, or birds twittering, and seem to follow on from transients like a foot fall frome somebody walking accross the floor, or from a sudden sound from an instrument like a violin. Additionally using Multiband Noise Gating removes these effects, but even with the settings set to remove noise only at a very low background level eg noise gate level -90db, the sound can be ruined. An example of this is a violin start to play a string. The initial start of the sound is eliminated, and the violin sound begins late and at relively high level.

 

I have now tried to use these techniques os several recordings, but the result is always the same. The recordings become unusable, and the recording with the initial background noise is always preferable. If the spectral subtraction did not introduce the new effects, there would be no problem. By the way, I have always used the smallest sample possible for sampling the noise to be remove, minimising the effects from noises additional to the background hiss.

 

Has anybody got any suggestions?

[musikone]

Using spectral subtraction on recordings to remove hiss, hiss removal is effective. However additional sounds seem to be added. The sound a bit like crinkling of silver paper, or birds twittering, and seem to follow on from transients like a foot fall frome somebody walking accross the floor, or from a sudden sound from an instrument like a violin. Additionally using Multiband Noise Gating removes these effects, but even with the settings set to remove noise only at a very low background level eg noise gate level -90db, the sound can be ruined. An example of this is a violin start to play a string. The initial start of the sound is eliminated, and the violin sound begins late and at relively high level.

 

I have now tried to use these techniques os several recordings, but the result is always the same. The recordings become unusable, and the recording with the initial background noise is always preferable. If the spectral subtraction did not introduce the new effects, there would be no problem. By the way, I have always used the smallest sample possible for sampling the noise to be remove, minimising the effects from noises additional to the background hiss.

 

Has anybody got any suggestions?

===================================================

I have never experienced this particular problem with spectral subtraction, and I use it regularly with consistently excellent results. NO trace of any problems, using Windows XP.

 

I cannot imagine why _you_ experience this problem when _I_ do not.

 

In short, this anybody has no suggestions for you.

 

 

Musikone

[MikeS]

I don't know if the effects on noise reduction may be something to do with the music being recorded. In passages with a consistently high level, as is often found in rock music, the effect is not very obvious. It is much more noticeable in music with load transients followed by quiet passages. The effect seems to follw the load transient, and may continue in the quiet passage if there are bumps (say from somebody moving a chair or coughing) when the effect is again apparent. I believe at least one other user has come across this effect and request a "soft knee" for noise reduction as it was creating effects similar to those caused by clipping (can't find this one again). Aso MarkR reported unwanted effects 20/02/07.

 

For Muskone's information, the effect is real and repeatable, but the degree of the effect is dependant on the source material.

 

 

MikeS

[musikone]

I don't know if the effects on noise reduction may be something to do with the music being recorded. In passages with a consistently high level, as is often found in rock music, the effect is not very obvious. It is much more noticeable in music with load transients followed by quiet passages. The effect seems to follw the load transient, and may continue in the quiet passage if there are bumps (say from somebody moving a chair or coughing) when the effect is again apparent. I believe at least one other user has come across this effect and request a "soft knee" for noise reduction as it was creating effects similar to those caused by clipping (can't find this one again). Aso MarkR reported unwanted effects 20/02/07.

 

For Muskone's information, the effect is real and repeatable, but the degree of the effect is dependant on the source material.

 

 

MikeS

=============================================

The type of "noise" which is reduced by spectral subtraction is a hiss, such as is noticeable in quiet passages in classical music, where there is not supposed to be any sound. No sound shows up in the WavePad sound window as simply a straight horizontal line. Noise which is amenable to reduction by the spectral subtraction method shows up as an undefinable (except by a spectral analysis), relatively constant (in a collapsed time scale) amplitude signal. Amplifying it and listening to it reveals a "hiss." In other words, the type of transient to which you refer is _not_ going to be reduced by this method of noise reduction. So you are right in your contention that spectral noise reduction "does not work"--that is, on the type of "noise" that you believe it should reduce.

 

This is why spectral subtraction noise reduction works for me but does not work for you. An apple is an apple and an orange is an orange......

 

 

Musikone

[mike20021969]

i suggest uploading 2 samples of the source you are removing hiss from to zshare.net (about 1 minutes worth).

one that has not been processed,and the version with these artefacts on that you mention.

then we can see if we can recreate the problem you mention,or whether it is unique to you.

 

http://www.zshare.net/

 

also,this is one of the problems with this 'affordable' software.

to get high end performance effects,such as noise reduction,adobe audition might be more what you need but for this you need to pay a premium price.

[musikone]

i suggest uploading 2 samples of the source you are removing hiss from to zshare.net (about 1 minutes worth).

one that has not been processed,and the version with these artefacts on that you mention.

then we can see if we can recreate the problem you mention,or whether it is unique to you.

----------------------------------------------------------------------------------------------------------

I don't have any problem with the spectral noise reduction in the WavePad software. So what would be the purpose of my doing this? Perhaps this "problem" is unique to _you_ :-)

 

 

 

http://www.zshare.net/

 

also,this is one of the problems with this 'affordable' software.

to get high end performance effects,such as noise reduction,adobe audition might be more what you need but for this you need to pay a premium price.

------------------------------------------------------------

So why do _I_ need something like Adobe Audition?

 

 

Musikone

[mike20021969]

note for musikone:

what made you think i was replying to you? since when did you become the centre of the universe??? errr....like NEVER!

 

the reply was for MikeS.

he is the one who started this thread with the problem he was having.

[musikone]

note for musikone:

what made you think i was replying to you? since when did you become the centre of the universe??? errr....like NEVER!

 

 

---------------------------------------------------

I thought that you wanted to learn.

I apologize for the error.

 

 

Musikone

---------------------------------------------------

the reply was for MikeS.

he is the one who started this thread with the problem he was having.