Believe it or not, I compress some of my audio files to below 4Khz, I like to compress things to save as much space as possible because sometimes it's just a couple sinewaves at low frequencies. It won't play files sampled to 3999Hz or below, but will play anything above 4Khz. So for some of these I use the default xplayer that came with my linux mint installation, but that only plays down to 2Khz. I know this is extremely niche, but is there any way to set a lower samplerate?

I think aplay plays down to 2khz. You could try lower, but I think it's wisdom to avoid low frequencies in speakers. What you do with aplay I imagine can be done with arecord.

EDIT: Before you ask, I wouldn't like to listen to any of those sound files .

I prefer to use "samples/second" for sample rate and "kHz" for the audio signal, to avoid any confusion.
4000 samples/second can give sounds up to about 2000 Hz (with some distorsion, of course). 2000 Hz is perhaps enough for sinus waveforms with no or few harmonics, so in theory it isn't strange. I think that low sample rate is rather too low for usual music sounds and music content.

How do you compress the signal? do you use any mp3-like code? I think lame (Lame Aint an MP3 Encoder) doesn't go lower than 8000 samples/second and 8 kbps of bitrate. It is true that one can use non standard values through the command line, at least for bitrates higher than 320 kbps but I don't know for low values.

If it's about saving space wouldn't extreme compression be better? I.e. convert to something like 16kbps vorbis/opus/aac?
If that's not enough, downsample to 4kHz as well.
Although, I can't imagine anything I'd like to listen to that is so reduced.

Not obvious it is any regular audio signal.
I remember (35-40 years ago) in a university laboratory, people used to take physical mesure results and store them. They used analogue tape recorders (open reel Revox tape recorders). Because the frequencies to store were very low (from a fraction of Hz to 50-100 Hz max) they used frequency modulation to translate the spectrum to the bandwith the recorder could write to the magnetic tape. Listening to that recording would give something like the modem or fax sounds.
Now, this translation could be perhaps avoided using newer data formats.

I tried once to encode regular music to some very low bitrate mp3 using a mp3 encoding software (I don't remember if it was lame or perhaps cdex). At 8 or 12 kbps, the sound was like a very old bad telephone line!
It seemed however than for low bitrates aac+ format was somehow better than mp3 (I used it for web radio casting around 2010, with edcast). It gave better sound for the same bitrate.

Yes, and I think even ogg/vorbis is much better than mp3 for low bitrates.
And opus is supposed to be even better, at least for human voice.

They're all in FLAC and it's not scientific data.

If you search the source code for 4000 then this code comes up: https://git.videolan.org/?p=vlc.git;...ff;hb=HEAD#l82
Given the hard-coded value it seems unlikely there's an option to override it.

Tracing the change back shows the check was originally added in 2008, but the commit message just gives the unhelpful non-explanation of "Check against too low sample rate" without stating why 4000 is considered too low.

You could try asking on the VLC forums, but since a couple of people have asked the same question as you with no responses that's probably a waste of time - so I'd suggest reporting it as a bug (along with the reason why you want to go lower), and see if that gets you anywhere.

Does it have something to do with the audio buffer size of the playback loop? If the samplerate is too low doesn't it fill up too slowly? Also thanks for going to the trouble of checking the code like that!