Paste a raw podcast transcript and get a tidy version with clear speaker labels, paragraph breaks at each speaker change, and optional filler-word removal.
Speaker labels are normalised to a capitalised label on its own line, such as HOST:, with a paragraph break whenever the speaker changes.
Automatic transcription output is usually one dense block where speaker names appear inline in mixed styles. This formatter finds labels in the Name: form, rewrites each one as a capitalised label on its own line (for example HOST:), and starts a new paragraph every time the speaker changes, which makes the transcript readable as show notes or a blog post. Timestamps in [00:12:34] or 00:12:34 form can be removed, and an option strips the filler words um, uh and erm along with the stray punctuation they leave behind. The formatting is pattern-based and works best on transcripts that mark speakers with a name and a colon; it does not identify speakers in unlabelled text.
How should speakers be marked in the raw transcript? As a name followed by a colon, such as Host: or Priya: at the start of a turn. The formatter uppercases that label, puts it on its own line, and breaks the paragraph there. Transcripts with no speaker labels are returned as cleaned paragraphs without labels.
What does filler-word removal change? It deletes the standalone words um, uh and erm and tidies the spaces and punctuation around them. It does not rewrite sentences or remove other hesitations such as like or you know, so a light edit afterwards is still worthwhile.
Is the transcript uploaded anywhere? No. The formatting happens in the browser on this device, and the download is generated locally.