The native ffmpeg resampler is pretty awful: https://src.infinitewave.ca/
The "soxr" one seems to be good, but I don't see any reason not to simply use the best one in existence, which to my knowledge is izotope 64 bit SRC. After all we're concerned about preservation. Why unnecessarily introduce artifacts into the audio?
The reason the resampler makes a difference is because PCM does not actually contain the entire waveform at every point in time, only enough to be able to reconstruct it (near) perfectly according to the Nyquist theorem. So what a resampler does, simplifiedly speaking, is, it reconstructs the full waveform and then it measures new samples at the correct places of the new samplerate. If the reconstruction is not precise enough mathematically, you introduce errors/distortions that are then baked into the resampled audio.
Whether it's audible or not shouldn't even matter in my opinion because there's no reason to not just do it right when it's easy to do. Why accept a degradation of the audio if it's not necessary? But I have read accounts of people in the audio industry who say the resampler differences are audible. Someday I'll have to do an A/B test myself.
About the settings ... I personally just use the simplified one with "Highest Quality" since I'm not exactly an expert. I should note that I use it in Sound Forge, which it used to ship with.
The "soxr" one seems to be good, but I don't see any reason not to simply use the best one in existence, which to my knowledge is izotope 64 bit SRC. After all we're concerned about preservation. Why unnecessarily introduce artifacts into the audio?
The reason the resampler makes a difference is because PCM does not actually contain the entire waveform at every point in time, only enough to be able to reconstruct it (near) perfectly according to the Nyquist theorem. So what a resampler does, simplifiedly speaking, is, it reconstructs the full waveform and then it measures new samples at the correct places of the new samplerate. If the reconstruction is not precise enough mathematically, you introduce errors/distortions that are then baked into the resampled audio.
Whether it's audible or not shouldn't even matter in my opinion because there's no reason to not just do it right when it's easy to do. Why accept a degradation of the audio if it's not necessary? But I have read accounts of people in the audio industry who say the resampler differences are audible. Someday I'll have to do an A/B test myself.
About the settings ... I personally just use the simplified one with "Highest Quality" since I'm not exactly an expert. I should note that I use it in Sound Forge, which it used to ship with.