Turn episodes into show notes, pull quotes, and searchable text. Whisper AI runs 100% on your device — your unreleased episode never leaves your browser.
Files are processed locally and never uploaded.
Drop your episode audio (MP3, M4A, WAV) into the transcriber above.
Pick a model tier — larger is more accurate, smaller is faster.
Copy the transcript and shape it into show notes, quotes, or a blog post.
Audio is invisible to search engines. A transcript turns a 45-minute conversation into indexable text — Google can find the episode for the exact phrase your guest said, and listeners can skim before committing to a full listen. Shows that publish transcripts consistently see their back catalog keep working for them in search.
The second payoff is repurposing: one transcript becomes the show-notes page, five pull quotes for social, a newsletter section, and the raw material for a blog post. Recording is the expensive part; transcription multiplies what that hour of recording produces.
Don't publish the raw transcript as your show notes — spoken language reads terribly. Instead: lift the 3–5 key moments with timestamps (listeners love clickable chapters), pull one quotable line per segment, and write a 100-word summary in your own voice. The transcript is the quarry; the show notes are the sculpture.
Keep the full transcript on the episode page anyway, collapsed behind a “full transcript” toggle. It costs nothing, helps accessibility, and it's the part search engines actually index.
Whisper is remarkably good at clean speech and falls apart gracefully at the edges: heavy crosstalk, strong accents on niche jargon, and brand names it has never seen. Expect to fix proper nouns — guest names, product names, place names — on the first pass. Budget five minutes of cleanup per 30 minutes of audio and you'll rarely be surprised.
Two things Whisper won't do: it doesn't label speakers (you'll see one continuous text — add “HOST:” / “GUEST:” yourself at the turn points), and it won't preserve your umms and false starts, which is a mercy, not a bug.
This matters more for podcasters than for almost anyone else: transcription services that run in the cloud receive your audio — including embargoed interviews, unannounced guests, and takes you later cut. Here the Whisper model downloads once and runs entirely on your device; the episode file never leaves your browser.
That also means it works on a plane, in a studio with no guest wifi, or anywhere else you'd rather not route audio through a third party. Private by architecture, not by promise.
The practical limit is your device's memory — hour-long episodes transcribe fine on most modern laptops and phones, just slower. Longer files take proportionally longer; a 60-minute episode might take several minutes depending on the model tier you pick.
No — Whisper produces one continuous transcript without speaker labels. For interviews, add HOST:/GUEST: markers yourself at the turn points while you do your cleanup pass; it takes a few minutes and reads far better than auto-labels anyway.
MP3, M4A, WAV, and most common formats — whatever your podcast host or recorder exports. If your file is video (a recorded video podcast), extract the audio first or use the video-to-gif tool's sibling pipeline; audio-only is faster to transcribe.
No. The Whisper model files download to your browser once, then everything — decoding, transcription, the transcript itself — runs locally on your device. Your audio never touches a server.