OK, I'm basically a bedroom hobbyist producer, but I am an IT guy by day. At work I was given a short wav file that was recorded outside, very poorly on a street. The noise floor is maybe half as loud as the garbled dialog I'm trying to isolate. Only a couple of sentences and phrases are intelligible. I have Izotope Rx 5 And Spectralayers 4 I think, but I am fairly new to both programs (I've used Rx on guitar noise in the past) but so far I'm not having much luck with the modules. I'd like to do as much noise reduction and dialog enhancement as possible then gate the rest of the noise out I guess, but all of the noise reduction I'm doing is leaving a lot of artifacts which is making trying to figure out the dialog a real challenge.

Anyone have any advice at all for extreme noise reduction or challenging vocal extraction? I can probably download any additional tools. I'm more of a post-punk guitar guy, not a forensic audio specialist. THANKS

 

Waves has a couple post production noise reducers aimed at film making which may serve your purposes, though I personally don't think much of any of them.

• WNS Noise Suppressor
• W43 Noise Reduction Plugin

Due to the random / varying nature of outside noise it is rather hard to isolate and remove it. It is not like an air conditioner which is pretty much constant throughout the recording; the kind of noise that most noise reducers are designed for. That is why most movies place microphones on the actors or do Automated Dialogue Replacement (ADR) with studio recorded dialogue post production.

 

If you are planning to record material in the future and you want to avoid all the hell that is trying to fix source material after recording it instead of getting shut of as much of the noise as possible at source then I can recommend some gear which will not cost you much as a fellow guitarist, bassist, vocals (not singing!), drum programmer for my own stuff and audio engineer, live sound and live sound tech when not doing my own stuff. Just let me know as honestly it makes life so much simpler even though it isn't at all as fun as getting a new delay pedal and rack mount effects, It more than pays off.

As for getting rid of so much noise that is around half the level in places as the wanted sound, It just is not really possible to achieve without a slew of artifacts left as a result and even then it is I imagine still not removing all the noise, This would be the case even with the best algorithms available by CEDAR Audio, Ltd. They only goes so far, The most useable results which I get with RX amongst others (but nothing better than RX and nothing matching CEDAR's stuff as it is bloody expensive as you can probably imagine) is by using multiple instances of the same tool/processor set to not be working too hard, So a lot of serial "stacking" to de-noise for example, I've found that leaves the least amount of audible oddness but unless it is low level back ground noise, Some 50 or 60 hertz hum and such then I still get the artifacts. iZotope's RX is as good as any other for the price/performance. If you live near a dealer of CEDAR's stuff then give it a try if possible (You might even be able to run the stuff that you are having a big headache with through the best they have if you shove it on a flash drive) but you still won't escape the side-effects unfortunately, You can have a look and search for examples by their stuff starting hear https://www.cedar-audio.com/
So yeah learning to utilize RX's stuff as best you can is probably your best bet along with making sure to kill the majority of unwanted crap before capturing in the first place.

I do understand the frustration if it's pieces which are not going to be re-creatable and the vibe and such was just right, It's sods law.

All the best

Dean

 

Use Edison's noise reduction tool. To prevent warbly artifacts in the vocals, use it lightly multiple times, instead of cutting out all the noise in one big heap.

 

Hey @spiggy Is this for a Corporate video project, or a highly confidential audio capture that your company wishes to transcribe into text?

If it doesn't put your job at risk, perhaps those of us with audio recovery tools such as Rx 5 etc. could help you out for free if you don't mind forwarding a copy. Perhaps the recording really is fubar, or maybe it's salvageable. It's difficult to say without hearing it.

 

I find the Groove3 tutorials are often quite good at demystifying audio tools...
https://www.groove3.com/izotope-training-video-tutorials/iZotope-RX-5-Explained

 

Spectral Editing is the way to go.

But you need to have a strategy, because audio restoration is a destructive procedure, that means the order in which you process stuff matters.

You could try to isolate some of the noise, build a loop out of it, reverse the phase and blend it with the original to cancel out the noise to some degree. I bet it won't work well because noise and wind and rumble and whatever is never constant, so it won't cancel out. But if the recording is fucked up already, you could try.

Another way woulkd be to isolate the voices and their harmonics with a tool and then turn the noise down and bring the voices back in.

PS: the 50/60 hz rumble removal is only necessary for analog recordings where a transformer was involed that gets its power from the plug in the wall run at 50/60 Hz AC.

 

Clever tactic in theory. But in practice this will only add more noise. The noise loop will be 99% decorrelated relative to the original audio.

I would keep trying with RX. Try out the Dialog mode, probably won't work, but worth a try. Then go to the spectral denoise. Learning a bit of the noise and trying to reduce it while listening to "just noise" checkbox checked, playing with threshhold and reduction amount. Your goal is to get it so you can hear as much garbage as possible, yet almost no voice information (it's circus clown balancing act, almost never get it perfect).When you get it about as best as you think it gets, leave it there, switch the checkbox off, and start tweaking the reduction envelope by progressively dividing it in halves and tweaking the amount of reduction in each frequency range. This lets you strategically muffle the artifacts in certain frequency spots, letting more original noise poke out, but hopefully getting you more intelligible voice.

When you're done with this nightmare, try the Zynaptique Unchirp (artifact suppressor, high end regenerator) and Unveil (general "gunk" remover and reverb controller). Unfilter by same company can give you more voice clarity if the audio needs it. Good luck man!

 

Last edited: Mar 25, 2017

Well that's basically what I said in the next sentence: "I bet it won't work well because noise and wind and rumble and whatever is never constant, so it won't cancel out" ;)

 

Magnets... use more magnets

 

In SpectraLayers, select the frequency resolution that matches the most your spectral voice data, use overlap x8 or x16, then use the harmonic selection tool, and carefully select voice parts (you can hit TAB to preview the selection). Once you have a gross selection that has most of the voice data ok, copy and paste that to a new layer. Then mute the source layer, and clean carefully the new layer using the eraser tool, or copy some additionnal data from the source layer using the brush selection or lasso selection tool.
I can give it a try if you want.