Not during the recording — after it
The browser's speech recognition cannot tell voices apart, so during recording there are no speaker labels at all. Writing everything down as "Speaker 1" would produce a transcript that is entirely wrong later.
When you stop, the saved audio is processed in a batch and split by voice. Telling it how many speakers there are makes it more accurate; leaving the range open lets it invent a speaker out of background noise.
What matters is what happens when it is wrong
Automatic separation is not perfect. One person gets split in two, or two people get merged into one. So the way to fix it has to be part of the feature.
- Rename — correcting one speaker's name updates every line they said
- Move a single utterance to a different speaker
- From this utterance onwards — limited to lines by the same speaker
- Split one speaker in two, or merge two into one
"Everything onwards" that also overwrote other speakers' assignments would be destruction rather than correction. So the range is always confined to one speaker, and every option shows how many lines it will touch.
The result
- Kim SeoyeonProduct lead
- Park JunhoEngineering
- Alex WhitmoreInternational sales
- Kim SeoyeonProduct lead0:04
3분기 로드맵 확정 전에 두 가지만 정리하고 갈게요. 온보딩 이탈이랑 해외 결제 건입니다.
- Park JunhoEngineering0:21
온보딩은 지난주에 로그를 다시 봤는데, 가입 후 첫 녹음까지 도달하는 비율이 38%였습니다.
- Park JunhoEngineering0:39
이탈이 몰리는 지점이 워크스페이스 만들기 화면이에요. 개인 계정으로 바로 쓸 수 있다는 걸 아무도 모릅니다.
- Kim SeoyeonProduct lead1:02
그 화면을 건너뛸 수 있게 하죠. 조직을 만드는 건 나중에 필요해질 때 권하면 됩니다.
- Park JunhoEngineering1:18
가능합니다. 가입 시 개인 워크스페이스를 자동으로 하나 만들어 두고, 조직 생성은 대시보드에 배너로 빼면 돼요.
- Kim SeoyeonProduct lead1:41
그럼 그렇게 갑시다. 준호님이 이번 스프린트에 넣어 주세요.
…and 8 more turns follow.
A sample written to demonstrate the product, not a real customer meeting.
Common questions
- What if I do not know how many speakers there are?
- Leave it on "automatic" and it estimates. When you do know, saying so is much more accurate — an open range sometimes turns a cough or a door into a speaker.
- Does it work for online meetings?
- If the microphone is picking the other participants back up out of the speakers, the voices are mixed and accuracy suffers. Capturing system audio directly with the desktop app is much cleaner.
Want to try it yourself?
Personal accounts get 40 minutes free every month. Start without a card and buy minutes only when you need more.