what's in the export
- long-form draft prose — not a transcript with timestamps every line, not a paraphrase. a continuous document with paragraph breaks in the right places and a clean reading rhythm.
- paragraph breaks at speaker changes for interview audio. each new speaker starts a new paragraph, with attribution preserved on first reference and elided on follow-ups (configurable).
- paragraph breaks at topic shifts for dictation audio (single speaker). the model uses pause structure, sentence-opening cues ("so, the next thing is…", "okay, moving on…"), and topic-shift detection to pick break points. you adjust before export.
- filler words trimmed — "um," "uh," "like," "you know," "i mean," and the half-finished restarts speakers do mid-sentence. the wording underneath is preserved verbatim. nothing is paraphrased.
- timestamp callouts kept as editorial reference in the form
[00:14:32]inline at paragraph starts. these are scaffolding for the writer — click to play that second of audio and verify a quote, then strip the timestamps before publishing. (one checkbox: "strip timestamps on next export.") - attribution lines preserved — for interviews, "X said" / "according to X" framing stays where it appears in the source audio, so the writer can decide which attribution lines to keep and which to convert into block quotes.
- markdown and .docx exports from the same source. markdown for ghost, substack, hugo, and any static-site workflow. .docx for writers who draft in word or google docs (open the .docx in google docs with no formatting loss).
sample output
a 60-minute recorded interview produces a blog-draft.md that opens roughly like this:
# draft — interview with maria chen
[00:00:42] maria chen has been running supply-chain forecasting
at her firm for six years. she said the hardest part of the job
isn't the math — it's getting the operations team to trust the
numbers when they contradict shop-floor intuition.
[00:02:18] "the model was right and the foreman was wrong, and
nobody wanted to be the one to say that out loud," maria said.
"so we built a dashboard that made the foreman's intuition
visible alongside the forecast. once both numbers were on the
same screen, the conversation got easier."
[00:05:04] she traced the shift to a single quarter in 2023 when
a forecasting miss cost the firm a major retail account. the
post-mortem produced two changes: a weekly forecast-vs-actual
review with operations in the room, and a confidence band on
every projection.
[00:09:21] "people accept being wrong inside a confidence band,"
she said. "they don't accept being wrong against a single
number."
.docx export carries the same paragraph structure with word's standard styles applied: heading 1, body, block quote. drop into google docs or word and start editing.
the workflow
- record the interview or dictate the draft. phone voice memo, zoom recording, riverside, or a usb mic into your laptop. mp3, m4a, wav, mp4 audio — anything ffmpeg reads.
- upload and transcribe. a 60-minute interview is ready in about three minutes.
- run the draft pass. one click. the AI-tighten step removes filler, breaks runs of speech into paragraphs, and inserts timestamp callouts. you see a diff against the raw transcript and accept or reject paragraph by paragraph.
- review and edit in the editor. click any sentence to play that second of audio. fix any paragraph break that landed in the wrong place. mark sentences as block quotes if they should be pulled.
- export blog-draft.md or blog-draft.docx. keep timestamps in for now (writer reference), or strip them on export when the draft is ready for an editor.
when to use this vs. other exports
- blog post draft (this page) — long-form prose starter for a writer who will edit it into a finished piece. assumes substantial editing downstream.
- clean-quotes export — when the deliverable is verbatim quotes for a journalism piece, not a long-form prose draft. lighter touch on cleanup, footnoted timestamps for verification.
- show-notes export — when the same interview audio is also being published as a podcast episode and needs accompanying notes.
- raw transcript — when you want every word, including filler, with timestamps on every line. for legal, academic, or verification workflows.
privacy
for embargoed interviews, source-protected reporting, and unpublished book manuscripts dictated aloud, run the file in private mode. the blog-post draft export works identically; the audio and draft stay on your laptop.
pricing for blog-post draft export
$0.25 per minute, all exports included — including this one, plus the raw transcript, captions, show notes, and clean quotes from the same audio if you need them. no per-format upcharge, no subscription, no minimum. waitlist signups get the first month free and 50% off forever.