The sample rate of a file is a delivery decision more often than a quality one. Video wants 48 kHz, music streaming wants 44.1 kHz, and most speech models want 16 kHz mono before they will look at a recording.
Rate and Nyquist
A file recorded at 48 kHz can represent frequencies up to 24 kHz, half its rate. That limit is the Nyquist frequency, and it moves with every rate change.
Dropping to 16 kHz means nothing above 8 kHz survives. On speech that is inaudible, because the intelligible band sits well below it. On music it takes the top off cymbals, strings, and room air, and the change is not subtle.
The output panel prints the Nyquist limit for the rate you picked, which is the number that tells you what the export can and cannot hold.
Filtering before the drop
Content above the new Nyquist limit does not vanish on its own. Left alone it folds back down and appears as tones with no harmonic relationship to the music, which is far more audible than the loss of the original content.
The resampler filters above the new limit first, then converts. This is why a downsampled file sounds duller rather than harsher.
Converting up
Raising a rate is refused rather than performed. Interpolating a 44.1 kHz file to 48 kHz produces more samples describing the same band-limited signal, adds nothing, and grows the file. Any player, encoder, or audio interface that needs 48 kHz does this in real time already.
If a piece of software rejects a 44.1 kHz file outright, converting the container, not the rate, is usually the actual fix.
Rate, channels, and file size
44.1 kHz stereo 16-bit WAV runs about 10 MB per minute. At 16 kHz mono the same minute is about 1.9 MB, and the audio is still comfortably above what a speech model needs.
That relationship only holds for WAV and FLAC, where every sample is stored. In MP3, AAC, and Opus the bitrate decides the size on its own, and lowering the rate changes what the encoder spends those bits on instead of how many it has.
Preparing speech for a model
Whisper and most other speech models resample to 16 kHz mono internally. Doing it first makes the upload smaller and removes a step where a stereo file with an empty channel can confuse a pipeline.
Pick 16 kHz, set the channels to mono, and export as WAV. FLAC is a reasonable alternative when the file has to travel and the pipeline reads it.