Background: The sound clip (original: 40 minutes long) has been repaired, to my full satisfaction by an amazing, highly experienced professional. Can provide his info if anyone is interested.
This is the first time I have had to restore any audio. Back in another incarnation, I programmed Interactive Voice Response (Phone) systems, and used Adobe Audition (at the time was known as CoolEdit), but only to crop and normalize. That is the full extent of my experience with audio software.
It bugs me that I had to get a hired gun, and became fascinated by the technology. I am a EE by training, did my share of Fouriers in college, but never after graduation.
The first thing I did was to get the 30 day evaluation of Sound Cleaner II and applied its most basic “instrument”: the Broadband Noise Filter.
On purpose, I applied the maximum Suppression and got the “wine glass sound” which I am told -in this forum- is evidence that I went too far. I guess the next step would be trial & error, trying a range of settings. Have no clue whether “Whitening” and “Adaptation Time” are relevant/desirable.
Problem Number 1: My Sound Cleaner evaluation expires in a few days.
For comparison (all samples in OGG format so they are small enough for the forum)
The first minute of your original:
Your “Sound Cleaner” with maximum suppression:
The noise has been very effectively reduced, but at the expense of noticeable damage to the remaining audio. In particular, the first burst of noise is just too much to handle and the remaining audio is badly distorted,
The Audacity “Noise Reduction” effect (new version for the soon to be released Audacity 2.1) with default settings:
Again, the initial burst of noise is too much to handle, but rather than mash the sound, the effect has more or less ignored it. For the remainder of the track, the noise reduction is reasonable, a little less than the maximum “Sound Cleaner”, but the sound quality is noticeably clearer with less metallic bubbly sounds (less “wine glass”).
I guess my next step would be to remove any scratches left (as many as possible) manually, correct? (IOW: There is no filter possible?)
By the way: Beyond that artifact/defect, there is a very noticeable sibilant “S” (attached). I am not sure if they were fully present in the original audioclip, or were introduced as a by-product of the applied filtering.
In any event: What would you do about those? (*)
TIA
-RFH
(*) Other than asking the speaker to take dictation classes
Listening to your “show” (more like showoff , I am must confess I am in awe.
A big doubt remains, however. Please keep in mind that at this stage I should be reading “Audacity for Dummies”. Is there any way I can replicate your results exactly? Did you just click on a magic button, commanding Audacity to “Reduce the Noise of this File”? I even downloaded the version that you mentioned (Audacity 2.1) and it seems to me that I am required to define (declare, learn, whatever) what I understand to be noise. Is that the way you did it?
if positive, the implication is that one must become really experienced at picking the noise, and from then on, the computer does the rest.
As a reference, before discovering Audacity, I acquired 2 applications:
(a) Sound Cleaner II (does not require a noise sample)
(b) iZotope RX 4 (does require a noise sample)
Yes. I just looked for one of the longer gaps between words and selected that to make the “noise profile”.
The (pre-release) manual entry for the new Noise Reduction effect is here: http://manual.audacityteam.org/man/Noise_Reduction
Not really - I just picked a gap that had no talking.
Play the selection listening to ensure it doesn’t contain traces of music/speech
Undo the Amplify
adjust or change selection as appropriate
repeat until satisfied you have a good noise sample
apply the noise reduction
I would love it if the get noise profile had a facility to preview the selection, suitably amplified, with a Preview button in the Step-1 box of the dialog
Additional tip:
I also check that I’m not removing too much audio by selecting a small section of the audio (once again add a temporary range label), apply the Noise Removal/Reduction to it and play a slightly larger selection of audio surrounding it, listening particularly carefully to the transition points at the beginning and end of where the test noise reduction was done.
If satisfactory,
then, tidy up and then apply the Noise Reduction for real,
else, it’s back to the beginning with selecting a fresh noise sample (also after a tidy).
, I am must confess I am in awe.