Choose the counting unit
Distinguish recording duration, conversation duration, speech-active duration, and duration with usable reference labels. Hold music, long silence, menus, and abandoned calls can make these totals very different. State which measure your model task needs rather than comparing source proposals only by the largest available hour count.
Count channels consistently
A one-hour two-channel call is not automatically two independent hours of conversation. If each channel is delivered separately, preserve the relationship and explain the counting convention. Overlapping speech may contribute to per-speaker totals differently from elapsed time, so describe that distinction when reporting annotation coverage.
Account for exclusions explicitly
Track corrupt files, missing sides, out-of-scope languages, privacy-review exclusions, and clips too short to support the task. Report whether totals describe raw source recordings, screened candidates, or reviewed material. Completed conversion or transcription does not establish that every hour meets the intended content and source requirements.
Make pilot estimates transparent
Explain how sample measurements were obtained and whether they can reasonably describe a larger proposed source. Avoid extrapolating a handpicked clean sample into an assurance about the whole collection. Send Fiund your counting rules, minimum useful duration, and review criteria so discussions use comparable units and acknowledge unresolved scope. For example, a thirty-minute file might contain a long queue and a short conversation. Report those components separately if the task needs speech-active material. If the sample is too small to estimate their distribution, state that limitation explicitly instead of multiplying the cleanest example into an unsupported estimate of a larger source.
Review checklist
- Define recording and speech-active hours
- Document channel counting conventions
- Separate raw, screened, and reviewed totals
- Record sampling limits on volume estimates