I have a series of videos of speakers at a college retirement dinner. The speakers were miked, but there was a terrible air-conditioner noise in the background. I'm going nuts trying to figure out how to remove this noise. It's particularly annoying during the softer-voiced speakers.
I've tried Apple Soundtrack, Izotope RX, and Audacity, but none have worked too well for me. The best results I've gotten have been from a high pass filter at 150 Hz, but while that removes the rumble of the air-conditioner, it doesn't do much for the higher-pitched whirring noise that remains. And if I apply a stronger frequency, the speakers' voices lose their bass and start to sound tinny (especially the men). I suppose a second filter might then be appropriate, but I don't know enough about audio restoration to know which one.
Haven't had much luck with the various programs' noise removal tools.
Would anyone be able to take a look at a short clip of the audio and give me some recommendations? I've uploaded a minute of one of the speakers. At the end of the clip is a 2 second section of just the air-conditioning noise. The file is an aif; if something else would be better, just let me know. Thank you to anyone who can help.
Results 1 to 12 of 12
This was done with Adobe Audition3, using ONLY the broadband NR processing (plus normalize to bring the levels up to notice). It was a quick & dirty and was pretty agressive with the processing, so you will hear digital artifacts, but the noise is WAY lower.
One captures a section of ONLY noise and uses it as a "noise profile". That is the base with which you compare your "signal+noise" so that you can match closely to the profile & remove as much as possible (without introducing TOO much artifacts).
A combination of this and some S-curve dynamics processing (using expander/noise-gating) should help a lot.
However, your recording needs the kind of improvement that can only be done by having better miking habits in the field. That version will never be "perfect".
Thanks so much. I don't have Adobe Audition, but I'll see what I can figure out. Though a lot of what you wrote is above my head, especially "s-curve dynamics processing (using expander/noise-gating)." But I'll try!
I completely agree with your last paragraph; however, these videos arrived "after the fact" for me, so all I can do now is try to improve what already exists.
If you look at a graph of Input to Output, a straight line diagonal shows a linear progression. IOW, -50dB in = -50dB out and -10dB in = -10dB out.
An S-curve modifies the dynamics such that the loud parts get lounder and the soft parts get softer, but this variation is GRADUAL. Example: -50dB in = -60dB out and -10dB in = -6dB out. The cutoff/transition point should be low enough to capture/boost the quieter sections of the speaker (though not everything) yet high enough to diminish the noise.
Also make use of a bandpass filter which matches the spoken voice range (~80Hz/120Hz -> ~4kHz/8kHz). Try these 3 & normalization in varying degrees and in possibly non-standard order (though I'd start with S-curve dyn -> Broadband profile NR -> BP filter -> Normalize).
These might have different names in Audacity, etc., but they're quite common, so you should be able to find equivalents. Use them sparingly/judiciously if you don't want to introduce too much artifacts.
Slightly more aggressive than Chris version, bandpass filter from 80Hz to 12kHz, dynamic processing between 80Hz and 8kHz, level normalization to -3dBFS.
Adobe Audition used (AFAIR some demo versions was available in past - they are limited but still functional) - Audacity is significantly less advanced but i assume comparable results can be achieved.
Thank you to both of you for your efforts.
I listened to both clips. Is the bottom line that this is a tradeoff? In other words, when you remove the A/C noise, you get that watery/mechanical/unnatural/compression background noise in return? I'm guessing that's the situation since both of your clips turned out that way. And that was the result because this noise is pretty loud, correct?
If that's the case, then I think I may just stick with the improvement I was able to get with the 150 Hz setting in the high-band filter. It removed the rumble, just not the whirring--but at least it sounded natural and did make the speaker's voice a little clearer. Your improvements are amazing, but I'm just not personally a fan of how overprocessed the resulting sound became.
Generally described Noise Reduction works on assumption that noise is constant and relatively uniform. In your case noise is modulated (fan blades ?)- this may give variable spectral noise, side to this background noise (various sounds, reflections etc) that may be covered by noise but they make noise no longer uniform - this is slightly like removing layer of dust from old painting - sometimes it must be done in series steps not at once.
In Audition there is a good option to estimate this threshold - instead remove noise select keep noise - search for point where noise will be no longer modulated by voices and tones (anything different than noise) - this will be your noise threshold.
Avoid also longer FFT's - long FFT have better frequency resolution but at a cost spatial (time) resolution - at some point for long FFT (for sure higher than 4096 points) this may be perceived as strange echo/reverb like frequency (tonal) artifact.
Perhaps i've used also to aggressive dynamics processing (compress/expand), try something less aggressive.
I also suggest to avoid taking noise profile to long as it may be something buried in noise but with tonal characteristic (i.e. not audible ring tone or beep) - this will be subtracted from whole sound but as it can be predicted it will affect whole recording.
With patient approach i believe you may have good result.
Bandpass filtering seem to be a good starting point - sometimes it is better to apply bandpass before noise removal - sometimes after - try both variants. Perhaps noise changes force to work on recording separated for parts - frequently noise color change in time especially on magnetic tapes but also in space (various environment changes - even air temperature).
Always search for pure noise part and align selection to zero (null) crossing - FFT (even windowed) don't like suddenly abrupt signals.
And as lame excuse - i have noisy fan behind my head currently (doing some measurements in Faraday chamber) and don't have headphones with me.
I would recommend a high and low pass filter and then some mild noise suppression (you do not want to go overboard and get a computer voice) and finally some 'AM radioshow' boost and perhaps some new acoustics with a mild reverb.
I like iZotope (RX4 and Nectar in this case).
Last edited by newpball; 2nd Jan 2015 at 15:23.
Wow, thanks everyone!
We did a few test runs using some very basic settings. At this point, we're just looking to clear up the noise as much as possible while avoiding any fake-sounding effects as a result. I'd rather keep some noise in the audio as long as it sounds natural.
1) Here is the original, again:
2) This one uses a high-band filter to remove the A/C rumble and some very minor noise reduction for the hiss:
3) High-band filter with heavier noise reduction:
4) High-band filter with even heavier noise reduction. This one approaches that unnatural sound:
I tried the low-pass filter but it didn't seem to have much of an effect. I do not understand what is meant by:
finally some 'AM radioshow' boost and perhaps some new acoustics with a mild reverb
I've been told in an unscientific poll of lay-people that #3 is probably the best, although #4 is the clearest. Please let me know your thoughts. Again, we are working with a limited skill set here and need to just get this task done as well as we are able to. I'd love to spend more time and learn more about audio work, but unfortunately, that's not really possible. Hope you all understand where I am coming from with that.
Hey, I am sorry for bumping this thread, but as I had the same problem, I want to share with you how I solved it. I had a chance to know some guys that work in air conditioning servicing and they said that they can help me. After 2 hours they did such a perfect work, that I was surprised because my provider and the service said that need to change it, it means a lot of money, but these guys made an excellent work without changing the A-C at all. So if you meet any problems with A/C I can share with you their contact. No need to pay more.