Audio extraction.
yt-dlp handles the media download path where supported.
podcasts
Uoink your podcasts too.
The thing you do with videos also works for long audio: RSS feed in, speaker-aware transcript out, local corpus ready for your model.
podcast corpusaudio-first
# Episode titlefeed, episode URL, duration, publish date## SpeakersSpeaker 1 / Speaker 2 / Speaker 3## Transcript[14:22] Speaker 2: the claim is...## Entitiescompanies, tools, people, topicsany RSS feed
Add the feed. Let the helper poll.
Drop in a public podcast RSS URL. Uoink can fetch new episodes, run transcription, and file them beside your video library.
Local Whisper.
Transcription runs on your machine. Choose speed or quality based on the model size.
Diarization.
WhisperX separates speakers as labels you can rename later.
who wins
For audio people who need receipts.
Journalists
Pull every factual claim a guest made and keep timestamp citations beside the transcript.
Founders
Map a competitor's interview circuit and search every mention of a product, market, or investor.
Researchers
Build a private guest-claim corpus across a niche, then ask your model to compare positions.
Creators
Study how long-form interviewers frame questions, interrupts, topic shifts, and sponsor reads.
Analysts
Search multiple shows for every mention of a company, product category, or technical term.
Students
Turn lectures and seminar feeds into searchable notes you can cite later.
the local Whisper trade
Private transcription costs time, not trust.
Expect roughly 10 to 15 minutes of compute per hour of audio on CPU, depending on model size and machine. That is the cost of keeping raw audio and transcripts local. Uoink runs the job in the background and files the corpus when it finishes.
podcast FAQ
Useful caveats, not fine print.
Does Uoink identify speakers by real name?
Diarization separates speakers as Speaker 1, Speaker 2, and so on. You can rename them later when you know who is speaking.
Does local Whisper need a GPU?
No. CPU transcription works, but larger models take longer. Apple Silicon and dedicated GPUs can speed it up when available.
Are podcast feeds uploaded to Uoink?
No. Feed polling, audio download, transcription, and corpus writing run from your machine.