Page 1 of 1

Best Noise Removal Settings?

Posted: Sat May 30, 2009 12:23 pm
by Jon3000
I want to reduce the background noise from a voice recording, while minimizing the quality-loss from the speech.

I read somewhere else that it's good to amplify the noise profile before sampling it, and that seems to be the case. But how much amplification?

And how about the other settings? I've tried putting frequency smoothing to max (1000 Hz) and attack/decay time to 0, and that seems to work alright, but is there better?

I know I could spend hours experimenting but was hoping someone else had a few tips for this.

Thanks. :idea:

Re: Best Noise Removal Settings?

Posted: Sat May 30, 2009 3:07 pm
by steve
Jon3000 wrote:I read somewhere else that it's good to amplify the noise profile before sampling it, and that seems to be the case. But how much amplification?
That should not really be necessary, but I discovered that it sometimes provides (subjectively) better results. Generally between nothing and +6dB.

Noise reduction is a fiddly suck-it-and-see type thing when it comes to getting the optimum results. I usually use settings in the region of:
12 to 32
120 to 150
0.05 to 0.2


I usually start with these numbers, then tweak it from there:

24
150
0.1

Re: Best Noise Removal Settings?

Posted: Sat May 30, 2009 9:17 pm
by Jon3000
Thanks for that.

How about a technical explanation of what frequency smoothing and attack/decay time are? I'm far from an expert on audio.

Re: Best Noise Removal Settings?

Posted: Sun May 31, 2009 4:09 am
by kozikowski
I can give you the fuzzy generalities.

The reduction number will do no reduction at all at 0 and the noise (say background hiss or buzz) goes down as you increase the value. You can go too far and the voices will start to sound mechanical, clipped, sterile, and "too perfect."

While you're juggling that effect, you might get gargly/bubbling/echoes in the sound. Gently increase Smoothing up from zero until the gargling just goes away. Too far (1000) and the noise will appear behind each spoken word, but not in between words giving you a pumping noise effect.

I've never used the Delay value.

The worse the damage, the more critical each value becomes. If the "noise" has a personality very close to the person speaking, you may find only one value of each adjustment that works at all, and it's also possible that the least show damage and most natural effect happens with a little noise left in the background.

Some adventurous video editor claimed he got good results from a very bad but critical track by applying noise reduction more than once.

Another unfortunate but way too common problem is people trying to apply noise reduction to garbage. By the time you realize you need noise reduction, it's too late.

I have a very difficult recording we made in one of our conference rooms that's almost impossible to clean up. I need to see if I can find a copy of that. We ended up abandoning Noise Reduction completely, rolled off the bass air conditioner rumble below 100 Hz and shipped it.

Koz