I have a recording from a dinner event and there are hundreds of sounds from utensils, which are extremely bad to listen to. Many of them happen during dialog.
What is the easiest way to get rid of them? Can I for example identify the first one and let the software remove all of them? If so, how to do it?
In a way yes, by selecting a typical utensils sound and then using Select > Select Similar…
But another approach would be to Unmix Noisy Speech, and then target those for removal directly in the noise layer, you won’t need to be as careful when removing them.
@Robin_Lobel I am so fresh to the software, I am having hard time even selecting the sound and using Select Similar feature. When I use Select Next, it keeps selecting areas without any sound just next to the originally selected area.
I tried Select All option and after a long while I got “a milion” areas selected. The app is barely responding. How can these be similar, if many of them are even empty.
it takes a while to understand what you are looking at in the spectrograph
I’ve never even used Select Similar and I spend lots of time editing similar sounds
yes, this is how I would do it too…then learn what those sounds look like in the noise layer
learning the selection tools is plenty of teething; learning the Replace, Add, Subtract and Intersect is key to the selection process
then you are in for a lot of manual removal …and while you are doing that, goodness knows what else you gonna wanna fix!
I did the unmix noisy speech before selecting as shown above. The selection is on the noise layer. I need to figure out how to select all of those noises based on one I select and remove them in bulk. There are hundreds of them, so one-by-one is not an option.
I do completely understand…as I’m manually reducing/ removing thousands on unwanteds in my big project…it can be a massive amount of work
how are you going about it? destructive editing? using erase…or delete commands?
Try to reduce the Similarity percentage.
I too have difficulties in getting ”Select similar” to work but when I do manage I set the Similarity to about 10-20%.
Also, try with only Match Noise or Match Transient ticked, if it’s a rhythmic sound you are about to select.
Maybe it’s easier if you perform an Unmix Components first and try Select similar on one (or two) of the resulting layers, e.g. the Transient layer.
That will be my next step to figure out after I select the correct sounds in the whole clip.
The answer was actually to increase the Similarity to 86%, as higher the similarity, the more closely the sounds must match to be selected.
After few attempts of selecting a sound and using the Select Similar option, I can tell that it is quite unpredictable, in terms of what similarity % is needed to only select the similar sounds, and in terms of being able to tell if the software is doing any intensive processing.
I got into a situation few times where there is no progress popup visible, while the SL application is doing something very intensive.
I am not able to replicate my previous success in selecting only the correct sounds…
The question here is actually what similiarity actually means in the spectral domain and what the parameters can accomplish.
Take this example:
That’s what I got:
So, what is this good for?
Yes, that would of course be correct in your case, sorry! 
When I have tried it I happened to get no similar selection at all (that is only the original) until I reduced the percentage. But that was not what you asked about, sorry.
So the higher percentage the fewer similar selections you get and vice versa.
No idea. My naive expectation was that it can select the same looking patterns across the timeline, at the same height in the graph. I am sure it is not an easy algorithmic process.
Select Similar also depends upon FFT size, so experiment with Select Similar percentage and FFT size to fine tune what gets selected.
It would be nice if somehow those functions and tools depending on FFT size could be marked to signal that to the user.
That is, FFT dependant tools could for example have a red icon and the others could be green. As an example.
@Sunnyman Select Similar looks horizontally, not vertically - as this would not make sense from a spectral/sound point of view.
What you’re looking for in your case is the Harmonics Selection tool, it has been made just for that : select simultaneously many harmonics.
Alternatively you can use the Frequency selection tool and select the fundamental, then click Select > Select Harmonics.
That would be nice, yes!
I would love a spinner icon shown in the UI (not a popup), that would be spinning when the app is processing. In Acoustica there is a progress bar built into the main window.
When using Select Similar, I have noticed that the popup progress bar does not always show up. It shows up when the process is longer, when the similarity % is lower. But if similarity % is higher, it never shows up, in which case I have no way to tell if searching has finished or not, apart from looking at the task manager in Windows and guessing based on the process CPU usage.
Noted !
