Talk Type · Episode 5 · 2 min ·
Who said what: speaker diarization
Speaker diarization is the feature that labels who's talking, turning a wall of text into a readable conversation. How it works, why it matters for interviews and meetings, and where it still struggles.
Transcript
This is Talk Type, from the team at Pepys, where we turn talk into text.
Picture a transcript of a four-person meeting with no speaker labels. It's just one long block of text, every voice mashed together, no way to tell who said what. Now picture the same transcript with each speaker on their own line. One of those you can actually use. The thing that makes the difference has a clunky name. Speaker diarization.
Diarization is the process of working out not just what was said, but who said it. The software listens to the voices, figures out how many distinct speakers there are, and tags each chunk of speech to a speaker. Speaker one, speaker two, and so on. You usually rename them afterward.
Why does it matter so much? Because most of the recordings worth transcribing have more than one person in them. Interviews. Meetings. Panels. Depositions. Podcasts. For all of those, who said it is half the information. In an interview, you need to separate the question from the answer. In a meeting, you need to know who agreed to what. Strip out the speaker labels and you lose the thread.
Here's where it gets hard, and it's worth being honest about. Diarization is genuinely difficult when people talk over each other. Two voices at once, and the software has to decide who's who in the overlap. It's also harder when voices sound similar, or when someone only says a few words the whole time. Crosstalk is the enemy. So the same advice that helps accuracy helps here too. One person talks at a time, mics placed well, and the labels come out much cleaner.
The payoff is a transcript that reads like a conversation instead of a monologue. You can skim it by speaker, jump to one person's answers, and quote them by name. For anyone working with interviews or meetings, that's not a small thing. It's what turns a recording into something you can actually follow.
That's this episode of Talk Type. The full write up, with the links and sources, is in the show notes. Pepys transcribes any file or link, any length, pay once, and we never train on your audio. Your first sixty minutes are free at pepys dot co. Thanks for listening, and we'll see you next time.