Move an audio file in time without touching the audio itself. A positive offset delays it, a negative offset pulls it earlier, and both are expressed as whole samples of silence added or removed at the front.
Shift by samples or milliseconds
The two controls are the same value in different units. Milliseconds are what a spec sheet quotes; samples are what actually gets written, and the sample field updates as soon as the file’s rate is known.
At 48 kHz one millisecond is 48 samples. Shifts under about 0.02 ms round to nothing and the file is copied through unchanged.
Measure against a reference
Load the take the file should line up with, and the offset between them is read by cross correlating the first few seconds of each. Leading silence is skipped on both sides before the search runs, so the measurement survives two files that were trimmed differently at the front.
The result is reported as how much later the loaded file’s content arrives than the reference. Use this offset applies the correction, which is the same number with the sign flipped.
Two readings decide whether to trust it. Correlation above 0.7 means the peak is unambiguous. A polarity note means the two files match but one of them is inverted, which the aligner does not fix.
What the export writes
A delay becomes a block of silence at the output’s sample rate and channel layout, followed by the file. An advance becomes the file starting at a later timestamp. Neither path resamples, and neither applies the short seam fade that joining two unrelated files needs, because there is no seam between silence and a file that was always going to start there.
The output length changes with the shift. Delaying by 500 ms makes the file 500 ms longer; advancing by 500 ms makes it 500 ms shorter.
Sub-sample offsets
Real misalignment is rarely a whole number of samples. A microphone 34 cm further from the source than another is about 47.6 samples behind it at 48 kHz, and the 0.6 is not recoverable without resampling the whole file.
That fraction is left in place. For phase alignment between two mics it is the part that matters, and correcting it belongs to an interpolating filter rather than a file shifter. For lining up a re-amped DI, an encoder delay, or two versions of one master, whole samples are the whole problem.