White Noise Suppression causes clicks

Discussion forum for Amadeus users

Moderator: Martin Hairer

Post Reply
Usabell
Posts: 12
Joined: Thu Dec 16, 2010 7:50 pm

White Noise Suppression causes clicks

Post by Usabell »

Since my other thread got hijacked multiple times and it seems that my answer there was somehow too hidden/lost or something, and no reply to the actual topic was made for a month now, I am going to start a new thread that covers the only single problem left. Thanks a lot for fixing the clipping problem with the White Noise suppression in the other thread! It definitely fixes the clipping issue the old version had, but still keeps the problems with the first few samples of a sound.

You can download these sound files to see the effects here:

http://audio.daniellippert.ch/HAUEP_D4_D4_E4_41_45.wav

http://audio.daniellippert.ch/HAUEP_F2_ ... 06_110.wav

On both of these sounds, after processing with -45 dB of White Noise Suppression applied, the waveform does not start at zero (vertical middle) on the very first sample (waveform starts without a zero crossing), like it did with the unprocessed sound files. You need to zoom in at the beginning to see this, and it can be clearly heard as a "tick" or "click" at the beginning portion of the sound. It seems that the noise suppression algorithm generates a click out of nowhere, if you watch the difference between unprocessed and processed sound. You can see what I am talking about here:

http://audio.daniellippert.ch/showdiff.png

The pink curve in this picture shows the sound wave after the processing, the dark blue curve is the curve of the original WAV file.

I really hope this can be fixed because I already started to work on this project more than a month ago and I need denoised non-clicking versions of these two and over 1000 other sounds.

Thanks in advance for looking into this!

User avatar
Martin Hairer
Site Admin
Posts: 1975
Joined: Wed Nov 08, 2006 11:49 am
Contact:

White Noise Suppression causes clicks

Post by Martin Hairer »

I am afraid that this kind of artefact is pretty much unavoidable.
You can easily create a batch processor that adds a "Fade In"
on the first couple of milliseconds, which would fix the problem.
Regards,

Martin

HairerSoft
http://www.hairersoft.com/



_______________________________________________
Amadeus forum mailing list
Unsubscribe / change settings at http://two.pairlist.net/mailman/listinfo/forum_list

Usabell
Posts: 12
Joined: Thu Dec 16, 2010 7:50 pm

Post by Usabell »

Oh, I am really sorry to hear that it can't be avoided. Funnily, Amadeus' White Noise Suppressor works much better than any other noise reduction algorithms I tried on this sound material, but the generated click changes the attack portion of the sound too much to be usable as a final product. I already tried the fade in trick (which was one of my first attempts to quickly fix this behavior, to be honest) but it somehow dampened the attack too much, even at very short fade-in times, so none of that worked. I am no DSP programming expert, but I wonder how a noise suppression algorithm creates such a harsh transient out of an almost silent portion of sound. Well, anyway, thanks for looking into this, I have to try out different solutions now and see which one has the least loss for my (important) transient material. I think the support for Amadeus is really great and I'll definitely continue to recommend it to all the people I know. Keep it up! :-)

EDIT: Looking one last time at the combined screenshots I posted I was just wondering if it wouldn't be an easy and also effective solution to, like, crossfade the unprocessed first, say, 30-40 samples with the processed sample material. It would definitely cancel out any unwanted clicking artifacts at the beginning of the sound, and I don't think any noise during the first 40 samples would be noticable to the human ear, so no loss here, either. The crossfade would have to be rather quick during the last 10 or so samples of the first 40 samples, and all samples before that could be taken over 1:1 from the unprocessed original. Well, just an idea, you decide :-)

CDJonah_alt
Posts: 379
Joined: Thu Mar 15, 2007 3:57 pm

White Noise Suppression causes clicks

Post by CDJonah_alt »

Can you add something to the beginning of the file so that the click is
outside the region where you need data?

On 1/26/11 7:46 PM, Usabell wrote:
Oh, I am really sorry to hear that it can't be avoided. Funnily, Amadeus' White Noise Suppressor works much better than any other noise reduction algorithms I tried on this sound material, but the generated click changes the attack portion of the sound too much to be usable as a final product. I already tried the fade in trick (which was one of my first attempts to quickly fix this behavior, to be honest) but it somehow dampened the attack too much, even at very short fade-in times, so none of that worked. I am no DSP programming expert, but I wonder how a noise suppression algorithm creates such a harsh transient out of an almost silent portion of sound. Well, anyway, thanks for looking into this, I have to try out different solutions now and see which one has the least loss for my (important) transient material. I think the support for Amadeus is really great and I'll definitely continue to recommend it to all the people I know. Keep it up! :-)




_______________________________________________
Amadeus forum mailing list
Unsubscribe / change settings at http://two.pairlist.net/mailman/listinfo/forum_list
_______________________________________________
Amadeus forum mailing list
Unsubscribe / change settings at http://two.pairlist.net/mailman/listinfo/forum_list

Usabell
Posts: 12
Joined: Thu Dec 16, 2010 7:50 pm

Post by Usabell »

An interesting idea! I just added 0.1 seconds of silence before the sound, and it resulted in a very different and at the same time interesting waveform. Instead of a sharp clicky attack it became a jittery, slowly rolling in shape before the actual sound starts. Somehow the noise suppression seems to look forward into future samples and generates some things depending on what it sees coming in the future. Maybe I can make a batch function that adds silence to the beginning, then processes the sound, then cuts the silence and fades in the very first bit, but the aforementioned fading method in my edit would be even better, even though I now doubt it will work because of the much longer artifacts it creates with the added silence at the beginning. The amount of samples that are altered from the "normal" waveform seems to depend on the actual waveform and the amount of silence in front of the sound, so it will be hard to predict the exact number of samples where the crossfading needs to take place. :(

CDJonah_alt
Posts: 379
Joined: Thu Mar 15, 2007 3:57 pm

White Noise Suppression causes clicks

Post by CDJonah_alt »

As far as I know, unless one does some very clever things that basically
turns on the filtering as you go through, you always get what are called
endpoint problems. As you note, it looks ahead, because of course, it
can't look behind at the beginning. I would try adding something like a
second at either end and then see if you can work out a batch process to
delete a second at either end.

Chuck

On 1/26/11 8:53 PM, Usabell wrote:
An interesting idea! I just added 0.1 seconds of silence before the sound, and it resulted in a very different and at the same time interesting waveform. Instead of a sharp clicky attack it became a jittery, slowly rolling in shape before the actual sound starts. Somehow the noise suppression seems to look forward into future samples and generates some things depending on what it sees coming in the future. Maybe I can make a batch function that adds silence to the beginning, then processes the sound, then cuts the silence and fades in the very first bit, but the aforementioned fading method in my edit would be even better, even though I now doubt it will work because of the much longer artifacts it creates with the added silence at the beginning. The amount of samples that are altered from the "normal" waveform seems to depend on the actual waveform and the amount of silence in front of the sound, so it will be hard to predict the exact number of samples where the crossfadi
ng
needs to take place. :(




_______________________________________________
Amadeus forum mailing list
Unsubscribe / change settings at http://two.pairlist.net/mailman/listinfo/forum_list
_______________________________________________
Amadeus forum mailing list
Unsubscribe / change settings at http://two.pairlist.net/mailman/listinfo/forum_list

Post Reply