transcribe-an-interview

# transcribe-an-interview Create a searchable representation that never hides what the recording cannot establish. ## Procedure 1. Preserve the source and record file identifier, date, duration, language, channels, and checksum. 2. Choose verbatim or edited-verbatim scope and define notation. 3. Identify speakers only from reliable production evidence. 4. Transcribe words, meaningful pauses, crosstalk, non-speech sound, and language changes. 5. Insert regular timecodes and exact markers for quotes, uncertainty, or source switches. 6. Mark `[inaudible]`, `[unclear]`, and phonetic guesses without inventing text. 7. Verify names, terminology, numbers, and quotations against audio and authorized references. 8. Proofread while listening at normal speed and spot-check a second time. 9. Save versioned transcript and correction history with access controls. ## Guardrails - Never fabricate inaudible speech from context. - Do not identify an unknown speaker by guesswork. - Protect sensitive interview content and machine-transcription uploads. ## Done - A timecoded transcript document records source, speakers, words, sounds, and uncertainty - Names, numbers, quotations, and sampled time ranges are verified - Corrections and source provenance are recorded