What software are you using for this? I don't recognize the screenshot?

It’s actually a Microsoft/IBM thing, released in 1991; WAV is just the file extension (and internal marker) of the RIFF container format when used to hold audio.

RIFF can hold many things: audio (WAV), video (AVI), MIDI (RMI), and Google’s new WebP picture format also uses RIFF as container. You can imagine this container like a tar or zip file, which also contain data on the inside, and some meta information about what these data are.

WAV files are RIFF audio containers; WAV has support for codecs, so it’s possible to put MP3-encoded audio into a WAV file (I’ve done that back in Windows 9x times so that the built-in Sound Recorder could play them; WAV4MP3 is a freeware tool that can add the WAV “container” or remove it to make it a plain MP3 file).

However, normally, WAV files contain PCM data (linear pulse code modulation, mostly 11'025/22'050/44'100/48'000 Hz, mono or stereo, 8 or 16 bit per sample). This is what you want, because it’s lossless (and, in fact, not even compressed).

There’s another lossless audio format out there, FLAC, which does compress; FLAC files are still quite a lot larger than MP3 or Ogg Vorbis files, but massively smaller than uncompressed PCM WAV files. If you want to store or upload/download largish amounts of audio, I’d suggest converting to FLAC; converting back to PCM WAV will produce the exact same sample data points the original had (probably even the exact same bits and bytes).

MP3 and Vorbis (the audio codec most often used with the Ogg container format) are both lossy compression algorithms, which means they introduce artefacts into the audio stream in order to be able to compress better. They try hard (using research into how the ear and brain works) to make that inaudible, but they are early lossy formats, and so they’re not all too good at that at usual bitrates (I’m told 320kbps MP3 is okay, and 256kbps Vorbis probably).

However, when you have a soundfont, the samples are processed a lot by the synthesiser: preamplifying, something with envelopes, panning, hall, chorus, other effects, pitch bending, volume change, mixing with other samples when you have polyphonic music, etc. to then produce output (MuseScore can produce PCM WAV, MP3 and Ogg Vorbis). You can imagine that a small error (from lossy compression) in the initial sample is multiplied many times during that process.

Then, the final output from MuseScore is most often MP3 or Ogg Vorbis. Taking a lossily compressed sample, decompressing it, then recompressing it again is not safe. Unless you use the exact same parameters, codec, compression software, etc. (but sometimes even then), you have a compression that introduces loss at different places, so this is also some kind of error amplifying. Here, you have a lot of changes (volume, mixing, etc.) in between, so you’re basically guaranteed to have bad results when starting from a lossy sample.

Then, there’s the soundfont to consider. SF2 stores the samples as PCM internally (omitting the WAV container because its information is already present in the SF2), but SF3 uses Ogg containers with Vorbis compression, which is also lossy. Now if you take a sample from MP3 (lossy), decompress it to PCM then compress it with Vorbis to put it into an SF3 format soundfont, you’ve again got error amplification because you use multiple different lossy compression formats.

This is, by the way, also the reason why, when editing soundfonts, you will always want the SF2, not the SF3. If you need an SF3, convert the SF2 to it after editing, but keep the SF2 in case you, or someone else, wants to make further edits. (Decompressing an Ogg Vorbis sample, doing changes to it, then recompressing, will almost always also introduce audible artefacts; hah, what an alliteration!)