Choose the time origin
Specify whether timestamps are measured from the original recording, a trimmed call segment, or the start of a channel-specific derivative. Include units and precision in the schema. A timestamp can be numerically valid yet point to the wrong speech when different artifacts silently use different starting points.
Track edits and offsets
Trimming silence, removing segments, concatenating legs, or replacing sensitive speech can alter timing relationships. Keep an explicit mapping to the source where required and permitted. Decide whether edited regions retain their original duration or whether a segment map must account for shifts introduced by the transformation.
Inspect alignment at several scales
Check call-level duration, turn boundaries, and representative word timings if word-level labels are required. Listen near interruptions, transfers, and long pauses. An opening segment that aligns correctly does not establish that timing remains accurate later in a recording, particularly after edits or conversion.
Define delivery checks early
Require linked identifiers for audio, transcript, speaker labels, and any alignment files. Record known timing limitations and acceptable tolerances for the task. Share these requirements with Fiund before sampling so potential sources and preparation plans can be assessed against the actual annotation and review workflow rather than a generic timestamp claim. Choose a transcript timestamp near the end of a trimmed recording and jump directly to that position. Verify the expected words and speaker against the audio. Repeat across a replaced or removed region if present. These spot checks can identify offset drift that a duration match at the whole-file level would not reveal.
Review checklist
- State origin, units, and precision
- Record edit offsets and segment maps
- Inspect early and late alignment
- Link every derivative to its audio identifier