Podcast and Interview Transcript Cleanup

Your auto-transcript is full of "um," "the the," misspelled guest names, and Speaker 1 answering their own questions. Most cleanup passes have the opposite problem: they paraphrase, so tentative opinions come out sounding certain, and on long episodes whole stretches quietly go missing. This skill cleans at a level you choose, changes only what that level allows, and flags every judgment call so you can check it against the audio.

What it does

• Cleans at a stated level (verbatim, clean verbatim, or lightly edited), so you know exactly how far the text has moved from the recording
• Removes filler, stutters, false starts, and back-channel "mm-hmm" turns, but keeps hedges like "I think" and "I want to say" that change what a claim means
• Builds a term sheet of names, products, and jargon, fixes speech-to-text errors only when context makes the fix near-certain, and marks the rest [unclear]
• Fixes speaker labels and catches diarization drift, such as a label switching mid-thought or a host who seems to answer their own question
• Flags ambiguous numbers, self-corrections, and can/can't mis-hears instead of picking one quietly
• Handles pull quotes (brackets and ellipses), SRT captions (cue timing preserved), and long episodes (labeled chunks, never summarized to fit)

What you get

The cleaned transcript with speaker labels and the source timestamps, followed by short editor notes listing corrections, assumptions, and timestamped spots to check against the audio.

Best for

Podcasters, YouTubers, interviewers, and content teams turning recorded conversations into transcripts, show-note quotes, captions, or source material for articles.