Problems with the new Noise Reduction replacing Noise Remova

[kozikowski]

Start recording. Hold your breath and don't move for two seconds. Breathe. Talk/Present normally for about 8 seconds. Cut off anything over 10 seconds total. Split to mono if you naturally produce stereo, Export WAV and post.

Scroll down from these forum text windows, Upload attachment.

As a fuzzy rule you should post clean, unprocessed work. If you post processed, we can't tell what the original was, what you did to it, or how to fix it. So that's limiting.

Koz

[kozikowski]

6 for Frequency smoothing is the maximum

Since 6 is about 150 in the old system, isn't it the case you can type in numbers up to 12 even though the slider doesn't go that far?

And no. I can clearly hear 0 and I complained about that.

Koz

[Gale Andrews]

6 for Frequency smoothing is the maximum

Since 6 is about 150 in the old system, isn't it the case you can type in numbers up to 12 even though the slider doesn't go that far?

Has Paul said that? Out-of-slider values are rarely documented.

The old Noise Removal went to 1000 Hz frequency smoothing which would seem to imply more than 40 bands equivalent.

And no. I can clearly hear 0 and I complained about that

As far as I know, "0" Frequency smoothing is 21.533 Hz at 44100 Hz sample rate, so that "6" is 150 Hz, and so "0" is not "off". Is that so?

But most people seem to be complaining that the default of 0 is too low, and that seems to be what Paul is saying. Get Sensitivity and Smoothing at the "best" setting, then you can take the Noise Reduction slider as high as you need without creating too much more damage.

Gale

[billw58]

As far as I know, "0" Frequency smoothing is 21.533 Hz at 44100 Hz sample rate

Zero frequency smoothing should be "off". If that is not the case then that is a bug.

most people seem to be complaining that the default of 0 is too low

I'm not one of them. What is your basis for "most people"? Given that frequency smoothing is meant to deal with extreme cases, I contend that the correct default value is zero.

- Bill

[kozikowski]

0 has not been my go-to value for a while now. After several false starts using 0 and creating trash, someone (I forget who) suggested greatly increasing the value based on their old tool equivalents.

I suggest Noise Reduction Of The Beast 6,6,6 should be default with a valuable variation being 12,6,6. I've used that multiple times now with various posters' difficult productions and high, sometimes startling success.

Given I'm working primarily with stand-alone voices and not music. Those settings could reduce music to garbage. I wouldn't know.

Koz

[kozikowski]

Last example in a long line. We're struggling with the poster creating very low volume recordings, and this is a step in the middle of testing. The patch does pass ACX Compliance and it sounds like the performer.

-74, -22, -3.2

The eventual goal of this posting is to not need a ton of processing to get the work out.

Attach 1: Submission
Attach 2: My patch
Posted: How I got there.

Koz



Peter Piper Processing
Audacity 2.1.0 on the supplied clip.
2015-04-29


-- Effect > Normalize: [X]Normalize to -3.2 [X]Remove DC > OK.

-- Steve's Rumble Filter.
Install and apply LF_rolloff_for_speech.xml
Effect > Equalization: LF_rolloff_for_speech, Length = 8191 > OK

-- Noise Removal.
Drag Select 1 sec to 4 sec. Effect > Noise Reduction > Profile
Effect > Noise Reduction > 12, 6, 6. > OK

-- Effect > Compressor: -20, -50, 3:1, 0.2, 1.0. > OK

-- Effect > Normalize: [X]Normalize to -3.2 [X]Remove DC. > OK


RMS (loudness) is a little odd in this clip. Drag-select just the spoken parts (9 through 20) and measure RMS. That passes. If you select the whole clip, the vast quantities of silence throw off the measurements. That's why we recommend two seconds of room tone on these tests and not the much longer one that ACX wants.

[kozikowski]

There's a mistake in that list. You need to select the whole clip before actually applying reduction. I don't think this will actually get used in its present form anyway. This is more a muscle-flexing, 'yes I can do it under duress' exercise.

I think there is something wrong with the poster's microphone. Even though bellowing into the microphone, the clip is very low. The presenter is soft-spoken, they said so, but still. They would have to resort to thermo-chemical explosions to get a -6 (good practices) sound meter reading.

Koz

[kozikowski]

Smoothing Text Entry.

I typed specific values into the text entry box before I found you couldn't do that and they stuck. The system did not reject them. I didn't know about the slider stopping at 6 until later. Since 6 works for me, I stopped worrying about it and went on to something else.

Koz

[Gale Andrews]

As far as I know, "0" Frequency smoothing is 21.533 Hz at 44100 Hz sample rate

Zero frequency smoothing should be "off". If that is not the case then that is a bug.

Hi Bill,

You wrote much of the Noise Reduction docs, but they do not accord with what Paul is saying about frequency smoothing. You say that frequency smoothing should be left at 0 except in "extraordinary circumstances". Paul is saying you need to turn smoothing up in most cases.

If each band is 21.533 Hz as Paul said, but 0 is off, how does six bands = 150 Hz? That seems like a docs confusion that we should resolve.

most people seem to be complaining that the default of 0 is too low

I'm not one of them. What is your basis for "most people"?

Robert J.H. is saying it, I'm saying it after playing more with Noise Reduction (though I don't use it for production work as you know).

Koz seems to be saying that he has not been using zero smoothing for a while on his speech examples.

It seems to be the case for Darren, according to what Paul says.

Given that frequency smoothing is meant to deal with extreme cases, I contend that the correct default value is zero.

So how can we resolve this when the effect's author is saying something else?

Time smoothing (not exposed) is hardcoded "on". Is that what we want?


Gale

[Gale Andrews]

I suggest Noise Reduction Of The Beast 6,6,6 should be default with a valuable variation being 12,6,6. I've used that multiple times now with various posters' difficult productions and high, sometimes startling success.

That would put frequency smoothing default at the right extreme of the scale, which looks weird but I agree that or somewhere between 2 and 6 usually gives better results than 0 (even with music).

Gale