Process transcript
You are processing a verbatim automatic transcription of an interview. The goal is to turn it into a clean, scannable reference document a writer can use as source material for a book.
Input: A raw transcript that may contain timecodes, speaker labels, filler words, and transcription artifacts.
Steps
1. Strip metadata. Remove all timecodes, speaker labels, and transcription artifacts.
2. Remove the interviewer. Delete the interviewer's questions. If a question provides essential context for understanding the response, fold that context into a heading or a brief bracketed note — e.g., [on his childhood in Odessa].
3. Clean the interviewee's speech. Remove filler words (um, uh, you know), false starts, and repetitions. Cut passages that are purely phatic or circular and carry no factual, emotional, or stylistic value. Preserve the speaker's voice: keep distinctive phrasing, dialect, unusual word choices, vivid digressions, and rough edges that reveal personality — even when they aren't strictly "on topic."
4. Structure with headings. Insert `##` headings and `###` subheadings that summarize the topic of each section. Headings should be short and descriptive, not interpretive.
5. Highlight key fragments. Bold the passages that carry the highest density of concrete facts, strong emotion, or striking language — the lines a writer would highlight with a marker. Aim for roughly 10–15% of the text to be bolded; if everything is highlighted, nothing is.
Guiding principles
- Cut aggressively for clarity, but err on the side of keeping too much rather than too little. The writer can always cut more; they can't recover what's deleted.
- The final text should read as a coherent first-person monologue, not a Q\&A.
- Output in Markdown.