I have some water drips recorded…using n 16db self noise xy cardioid. I have tried a lot of diff approaches to reduce noise but not get artefacting. Seem to be getting better results just gating it in Wavelab 
Does anyone do a lot of this and have top 5 tips for getting this done ie best practice? The YT vids are the same as what Im doing…
Any help greatly appreciated.
Cheers
This plugin might be helpful
https://www.kvraudio.com/forum/viewtopic.php?t=491777
Depending on the available ressources, a better microphone might help.
The price is probably out of range, but its results are … awesome
Okay, I listen to the file you provided.
The Lewitt LCT 540S will fit your needs perfectly
Is this noisefloor needed?
The denoising of SL is rather rudimentary, imho. It does create artifacts quickly. Therefore I use the mentioned plugin. I also use Wavelab, because the VST-interface of SL is just crab and practically unusable (realtime stutter step and no graphic animation when using Fabfilter/Melda plugins for instance), imho.
The VST-interface needs to be fixed and up to a level, where this is usable. We are in the year 2025 and not 2005.
It’s good, that the soundfile does contain a clear noisepart from 45s to 50s, which makes this easy to grab.
Thanks @Sunnyman
Appreciate the help…but question was related specifically to SL…it should be able to do this easily…if it cant di a better jib than spectral compression…not much point in owning it. I def dont want any more plugins…simplifying
I have lots of mics but specifically need a portable solution which is why i use it…and they absolutely need to be at least XY stereo…i have tlm 103, etc. And other 5db sn. Currently doing a binaural head
This mic setup has been fine for external field recording…its the quiet in the bath and isolating drips
Cheers
Ah ok.
I would recommend first remove the low noisefloor (0 - 200 Hz, maybe) a bit.
Than use the SL denoise (Register Noise). Select the part from 45s to 50s and play with it.
Nice drops, nice sunday.
Thanks
I thought there might be something i was missing…def did all that. I did a horiz freq range and trimmed that out as its not part of the sound (can you believe its a mower 100s metres away but the bathroom is a helmholz and as its in range…away it goes! I found it slightly better to do it and steps and iterate but yeah, either way it still gets the AI artifacts
Sunday is always play day…experiments always create unique things 
At some points of noise reduction artifacts bleed in. Use the reduction with care. And yes, a microphone with a lower self noise is the solution. Money= (Frequency Range)/(Self noise)
What you might also like is an ultrasonic recording of a simple key falling onto a plate.
Than playback 10 times slower.
This gives an amazing chime like sound play.
For binaural recording, did you try this?
When it comes to civilization noise, this is a pain. Cars, Mowers, Airplanes, …
Alle klar
I think i have a hybrid solution using denoise and gate
As with most of these solutions, they have terrible noise…that model is 23db
The issue is that the droplets are in different positions in soundfield and different sounds, some are large drops, some dropped from higher up and in diff pan postions. These will then be loaded as single shots into a kontakt instrument and used in the song which supports the subject if the lyrics. The pan postioning is much better from the xy vs static pan…hrtf would be ideal…just working woth what i have
Im actually making my own binaural solution..a composite unit and the capsules im.waiting on are 14db. Im actually an industrial designer and have worked for AEA…
Its a different design though…which incorporates a modular approach of all 3 types ie
On ear
Portable
Full head
Should have prototype in about 2 weeks
are you looking for an automated/module solution? quick turnaround? well, then need to get good at recognition of spectrogram and concrete a SL workflow.
reading the spectrogram takes some teething and plenty of practice
In my use cases, I typically reduce non-destructively rather than remove destructively and then re-balance if dealing with artifacts. Frankly, spot de-noising is such a better solution that the blanket procedures I’d been accustomed to just last year…(I’ve mostly been using Acon Restoration Suite and Extract Dialogue. I was using Sonnox Restore, too…but Acon had been “good enough”…and living with artifacts was a balancing act of what worked for the material. Again, I like to say wanted and unwanted noise.
Just because SL isn’t “working” for your use case doesn’t mean there’s no point to having SL for users other than yourself.
In my noise reduction and rebalancing tasks for documentary film work, SL is the best tool I have available, by a very large margin. Turnaround time can be slow and very time-consuming manual processing, but the result is jaw-dropping IME/O.
I would instead of Denoise modules use Unmix modules.
First I would run Unmix components.
Then I would use the Frequency range selection tool to mark the low frequency area where most of the noise reside. Maybe make the selection’s upper limit faded. And then push Delete. One of the layers is relatively free from this noise already (Transient).
Then, on each new layer I would use Unmix levels, and set the threshold values different for the three new layers. I would probably use the Peak Power variant and aim for a threshold value that would put as much wanted information as possible on the resulting High level layer without getting to much of that Low level info.
Then I would put the three Low level layers away and mute them and listen to the others before I would merge them.
This is a very quiet recording - it’s not easy to hear or see the noise, so maybe I would amplify it a couple of dBs before acting.
If so I would have to lower with the same amount afterwards.
Or I would have changed the display amplitude curves up in the right corner.
Interesting approach. I’ll try this next time.
I have cleaned up the noise for you in iZotope RX Advanced.
Removed the bump at the end, removed the LF noise.
Audio is then boosted by +18dB, from there made two versions with noise reduced by -12dB and noise reduced by -24dB.
Download the files here:
Sounds good.
I should have been really clear on OP
My focus is to prep the file for segmentation so that is can be played in kontakt/HPD20 andt will be metered as a played instrument with velocity modulation of parameters
Using the gear I have, the mics are actually decent quality and -18db noise is not great but no debilitating like many of the binaurals which are in the >22db zone unless you use the DPA capsules or go all the way and get the neumann…its not practical in context to get too carried away but I had assumed that with so many examples of SL doing amazing unmix and noise demos that this would not be a big issue and straightforward…well its not and hence the post
Yes and Of course…but I bought SL to do some “simple” things which so far, have not been great so I appreciate all the feedback here and hopefully helps others who are just making music. It has some excellent creative scope but I need the bread and butter to work first.
Even unmixing drums from stereo room mic is not great (I often do acoustic recordings world/jazz etc with a single R88 ribbon in ms or blumlein…would be nice to get just a tad more control with layers but that has not worked out so well)
That was my next stop…thanks for the direction…
Thanks that is helpful
Is that new in 11? I missed that not where that setting is
Very kind of you and the result sounds great but as mentioned, the point is to use less tools, in a better way than keep throwing more options at it. That is the issue…options paralysis and really just want less tools that I know better even if a bit clunkier
RE Workflows
Has been so great getting these persectives and really shaping handles on the tools
I have been a long time graphics person…so putting it all into perspective so far
I love the visual paradigm…but its just half baked and unusable without a lot of pain…just basic mask tools would be so helpful
ie
Smooth outline (with a linked time/freq and graphic referenc)
Feather etc also with a purely graphic reference when doing cookie cutting (or layer isolating in PS speak)
Remove Holes
Magnetic lasso
etc
shouldnt there be a amplitude value on the info bar?
Else how can you judge the parameters for levels? or am I missing something? Or do you have to have the sample dialog open all the time?
It’s always a challenge to work against the increase in entropy…
The microphones are decent and with low power consumption, but they introduce some noise.
Now, we use heavy computation on modern computers with ai-supported noise reduction algorithms, which consume - in comparison to the mic - a lot of energy just to decrease the noise floor some dB.
I like that and spent some time with it. The drops are fine, use echoes, reverbs, filter etc. Pure fun, using all the good stuff.
We live in truly amazing times.
What I meant by ’different threshold values’ was that each of the resulting three levels carries a different amount of noise.
The Transient layer has the least amount, if any.
The Tonal has some noise.
The Noise layer obviously has the most, where some of the noise actually is ’wanted noise’; e.g. the ’tails’ after each drop, and other ambience carrying information.
Hence the threshold values for the respektive layers must be set differently in the Unmixing levels dialogue box.
With a boosted audio file you could always use your ears to set the best value, but with this levelwise weak and quiet audio at hand it’s harder.
Like you show with your last screendump the mouse position shows only two of the dimensions of the spectrogram; time and frequency.
It would perhaps be of some help if it also showed the third aspect, the amplitude. If nothing else but for the sake of completeness.
The parameters for the Threshold value is set automatically after having done the Analysis in the dialogue box (Unmix levels). Unfortunately that value is not of much use since it will mostly be too high. You often, if not always, have to change to a much lower negative value for things to happen.
Maybe it’s easier to see and address the noise if you use the 3D-tilting of the spectrogram, the ’wanted’ areas would perhaps stand out a little more from the background/noise.
And using a colour scheme like Magma or inferno, where stronger amplitudes are whiteish.
Speaking of feathering selections, there is a new feathering (and also an unfeathering) tool among the tools to the left. It looks like a little sun, with a dashed circumscription. You can use that by clicking on the borders of a selection to make it more feathery.
Otherwise you can fade a selection - before using it - by setting the fade value in the information line to the left above the spectrogram. Applied on the Frequency range selection there will be fades on both the upper and lower border, but pushing the whole selection down towards the zero Hz the lower fade will finally disappear.
The Art of Noise(reduction)…
Wow!
When it comes to noise, it’ll never be perfect, because at some level in the data, noise is indistinguishable from actual value sound.
This is the key:
“The Noise layer obviously has the most, where some of the noise actually is ’wanted noise’; e.g. the ’tails’ after each drop, and other ambience carrying information.”
Ever wondered about the cocktail-party effect in real live and on a recording?
Is it possible to realize the cocktail-party effect in a recording? Maybe binaural? Actually, I doubt that.
Our mind filters the sound and eliminates the noise quite effective. And everyone does it slighty different (I believe). Than there is no general algorthm, that fits all.
So, in which way is noise different from the tails and fine details? What are the practical limits of noise reduction? And how far off we are from this limits today? I think, we’re pretty close, though.
Not that I ever look at those in real use case scenarios, I could see this being helpful
that said, doesn’t this show in the waveform window…if user has it visible? Nevertheless, user would have to zoom in the dB units strip very tight to get near to reading the actual value…turn on the grid…and a lot of navigational faffing to get accurate
I gotta say, flying by seat of pants after running Unmix Noisy Speech is my go to current workflow…as I’ve already described
And I just flat out don’t do any music in SL…just to check how unmix modules might have improved…so I haven’t checked that in a while
I have seldom used unmix components. I did take your advice on unmix levels last autumn, which does work well, but frankly, after getting to know the spectrogram better after much practice, I find manual tools work great…altho, as I often say, I do not use the “automated” tools that “magnetize” to amplitude (freq select, harm select, magic wand) due to paste anomalies I have previously posted about on this forum…all manual tools for me excpet running some modules
I would just find it very easy for eg Unmix levels if I could just hold my mouse on a threshold area and input that as the analyse does not work for me.
The waveform editor, as described…is of no real use for “real” editing in the old paradigm (for me). The unmix comps and a quick tweak did it and quite comparable the example by @Robert_Niessner (once again thanks Robert). I just exported and did the usual sample slicing in WL then drag and drop into kontakt.
Just getting a bit more perspective on tools really helped. For this example, in the future with better mask tools, I would just do a combination
I didnt do it on this one but I did experiment/trial and that definitely did the best job with no residual noise and min artifact at high monitor volume…more than good enough for this.
@ctreitzell Have you actually used that tool in serious PSD masking. it is using the graphic boundary (a threshold) and my context for the tool was not freq or amp but rather, as the SL paradigm goes, using it purely graphically. Doing this manually without magnetising (rather painful tbh) the outcome described above was the best I had heard. I only needed 12 drops from the recording. Maybe you have used it…but maybe your context says otherwise 
Some much great help here…thanks again
