Every episode, word for word.

Each episode was downloaded from the public podcast feed and transcribed locally with Whisper and pyannote. Speakers were matched to hosts by voice fingerprint across the whole archive.

• Not audio-verified. Words can be misheard, and Whisper tidies speech: it drops most “um”s, never writes laughter, and uses American spelling.
• Speaker labels are approximate. Quick back-and-forths are sometimes merged into one speaker, and a few of Gavin’s lines are labelled Geoff.
• Unidentified is anyone the voice map couldn’t place: ad reads, guests, clips and film audio during watchalongs.
• Timestamps mark where a paragraph starts, measured from the start of the published audio, ads included.

Updated recently. Spotted a mistake? The timestamp links make it easy to point to.