Audio Sample Rate: Check Resampling Before Trusting Annotation Times
· Annotation Quality
Changing a rate label can change playback duration. Verify the actual transformation before accepting audio annotation timestamps.
Changing an audio track's playback sample rate and resampling the audio are different operations. For an annotation delivery, compare the source and delivered sample rate, sample count, duration and timing references before accepting labels against the transformed audio.
This guide is for annotation buyers reviewing audio datasets. It does not advertise jobs, eligibility or pay. Data Annotation Companies is independent and is not affiliated with Audacity.
Why can the same samples have a different duration?
The Audacity Audio Track Dropdown Menu manual distinguishes its Rate operation from Resample. Raising the playback rate of the existing samples makes them play faster with a shorter duration and higher pitch. Resampling changes the sample representation while preserving the track's speed and length in the documented operation.
This is a tool-behavior distinction, not a recommendation to process every delivery with that tool. Confirm the actual application version and transformation. A file's new rate value alone does not prove which operation occurred, whether speech remains intelligible or whether the supplied labels still align.
What does a simple sample-count check show?
Illustrative example. A fictional mono recording has 220,500 sample frames at 22,050 frames per second. Its duration is 220,500 / 22,050 = 10 seconds. There is one sample per frame because this example has one channel. Do not use a stereo file's total individual channel-sample count as if it were its frame count.
| Case | Arithmetic |
|---|---|
| Source | 220,500 / 22,050 = 10 s |
| Same frames at a higher rate | 220,500 / 44,100 = 5 s |
| Duration-preserving resampling | 441,000 / 44,100 = 10 s |
The five-second result does not preserve the original ten-second time base. The final row describes the sample-count arithmetic for an assumed duration-preserving transformation; it is not proof of the actual output or its audio quality. Real formats, padding and processing boundaries require inspection of the delivered file and transformation record.
How can an unchanged sample index move an annotation?
In the same fictional source, frame position 44,100 corresponds to 44,100 / 22,050 = 2 seconds. After the assumed duration-preserving conversion to 44,100 frames per second, that time corresponds to frame position 88,200. Reusing the old index 44,100 against the new rate would refer to 1 second.
Keep an explicit zero-based position convention and the source asset version with the index. Labels stored in seconds need a different check from labels stored as frame positions. Even when duration is preserved, verify representative boundaries and any trimming, silence removal, channel changes or time offsets before claiming the labels remain aligned.
Which delivery fields make the check reproducible?
| Field | Record |
|---|---|
| Asset | Source and delivered IDs and versions |
| Rate | Actual frames per second for both files |
| Frames | Decoded count and channel basis |
| Duration | Measured values and comparison |
| Transform | Tool, version and actual operation |
| Labels | Seconds or frame indices, offsets and reviewed alignment |
Do not repair a mismatch by relabelling the metadata until the files appear consistent. Preserve the original, ask the delivery owner for the actual transformation and approve a corrected asset-label pair with its version. A matching duration is one check, not evidence that every word boundary or class label is correct.
Frequently asked questions
Does changing the displayed sample rate prove resampling happened?
No. Verify the operation, decoded frame count and resulting duration.
Can an old frame index be reused at a different rate?
Do not assume so. Reconcile the time base and approved mapping.
Does matching duration prove annotation accuracy?
No. Check the actual audio and label alignment under the agreed policy.
Use the pilot acceptance register to keep delivery counts and reviewed outcomes separate. The dataset labeling overview introduces broader file preparation; this worksheet addresses the specific audio time-base check. Browse the company directory and ask providers to demonstrate their actual audio transformation and alignment controls.