Count the samples in the example
Suppose an owned ten-second recording contains 480000 samples in one channel at 48000 samples per second. If the same samples are played at 44100 per second, they last about 10.884 seconds. That is a changed duration, not a duration-preserving delivery conversion.
A resampled ten-second version at 44100 samples per second instead represents the same interval with 441000 samples. This arithmetic is hypothetical and omits container-specific trimming details. It still makes the difference between changing interpretation and creating a new sample sequence clear.
FFmpeg documents asetrate separately from aresample. Similar-looking rate controls should therefore be checked against the intended task, using the installed version's help.
Trace the actual processing choice
If an export sounds lower and ends later, preserve both versions and inspect the processing record. Do not compensate first by shifting captions or cutting the final second. Those changes can hide the consequence without correcting the chosen operation.
Make a small candidate with the intended duration-preserving conversion, then compare voice pitch, a recognisable event and the ending. Keep the audio channel inventory fixed during the comparison so a different mix does not confuse the result.
Record the source rate, output rate and whether duration was intended to change. A later production can then distinguish an accidental reinterpretation from a deliberate creative speed adjustment. The accepted output serves the stated goal rather than merely displaying the requested rate.
Questions answered
Will unchanged samples have the same duration at a new rate?
No. Duration depends on the rate at which those samples are interpreted.
Should I shift captions to hide a rate-conversion mistake?
Correct the media operation first, then review the matched package.
Sources & further reading
Official references for the subjects discussed. Service terms and device software can change.