I don't know of any data sets that are about 'natural' noisy data (I bet they do exist) but you could try looking at some old archival data, such as those from NASA (https://www.nasa.gov/home/hqnews/2011/sep/HQ_11-321_NASA_Sounds.html) perhaps?

Ciao Ramiro. There is no lack of noisy data since all real data are noisy. I have tons of NMR data with various levels of noisiness; the only problem is that they come from real instruments in all kinds of binary formats. Tell me what format you need and I might write a small utility to export some of them.

I use intensively automatically adaptable Savitsky-Golay filters. Used as filters, B-splines are biased, I believe, which might be a heavy handicap (I have discarded them long time ago). In any case, if you dig these things, we might have work for you; get in touch.

Look, saying "some data" is not good enough :-) I suppose you need 1D arrays of real or complex numbers, not 2D, 3D, etc arrays. Maybe you want them sampled on a regular grid, or maybe not. To get them into R, you probably imagine ASCI columns in some text files that can be displayed and one can read them something like 12.3456. Which is logical enough, but not the way data are normally acquired and stored in real life (at least not in my world). It would make the files real huge and there is really no need to actually read any individual values by human eyes. In NMR, for example, a 1D spectrum is usually collected by an OEM's instrument into a folder with up to 12 files, one of which contains the actual data, but they are binary and when displayed in a plain text reader (like the Notepad) you see just gibberish. You need special visualization softwares to see something intelligible like the figure I am attaching (just the real part of a small portion of a spectrum). Now, if you wanted to bring that portion into R, I would have to write a small program to convert it into an ASCI format that YOU could read. You see the noise there and, of course, it is nothing 'acoustic' (a funny notion). I am pretty sure that almost any scientific data look somewhat like this, including those from various telescopes and many other scientific instruments. And always the first thing you have to face is the format in which the data are stored, versus the format you expect. Welcome to the real world :-)

Now, look again at the data. We need to reduce the noise as much as possible while distorting the peak shapes as little as possible. Maybe that makes clearer my comment about Bezier splines being "biased" (not a happy expression in this context); you probably already see the problem.