How to transcribe audio with multiple speakers (and know who said what)
Short answer
To transcribe several speakers, use a tool with speaker separation (diarization), which labels voices Speaker 1, Speaker 2 and so on. Rename each label once so names apply everywhere, and merge a speaker split into two. It works best with one person speaking at a time.
A transcript of a meeting or interview is only half useful if you can't tell who said what. Speaker separation (often called speaker diarization) splits the conversation by voice, so each line is attributed to a speaker you can then name.
Here's how it works in Dolcenota, how to tidy up the labels, and how to record so that every voice comes through clearly.
Get a transcript by speaker in 3 steps
- 1
Record or upload the conversation
Record in your browser or upload an audio or video file. Voices are separated automatically; there's nothing to set up.
- 2
Name each speaker
The transcript labels voices Speaker 1, Speaker 2 and so on. Rename them and the names apply everywhere: the transcript, the summary, action items and exports. When the conversation mentions someone's name, Dolcenota may suggest it; you confirm it with one tap, and it's never applied without you.
- 3
Fix a split speaker if needed
If one person was split into two labels (for example, after changing seats or microphones), merge them into one. It's reversible: you can undo the merge anytime.
How speaker separation works
The AI groups stretches of speech by the characteristics of each voice. It doesn't know who people are; it knows that the same voice spoke here and here. That's why the labels start as numbers and you add the names.
What makes it work better
- Distinct voices and one person speaking at a time.
- A microphone roughly equidistant from everyone, or one recording from the call itself.
- Little background noise and no music.
- Introductions at the start ("I'm Ana, from finance"), which also help with name suggestions.
Where it struggles (any tool)
- People talking over each other.
- Very short interjections ("yes", "right") that can be attributed to the wrong voice.
- Similar voices recorded from far away.
- Large groups where many people speak once or twice. Name the main speakers and leave the rest.
What you can do with a transcript by speaker
- Attribute quotes correctly in an article or report.
- See action items grouped by the person who committed to them.
- Ask "what did Carlos say about the budget?" and get an answer about that person.
- Export to Word, PDF or TXT with the speaker and timestamp on every line.
Frequently asked questions
How many speakers can it identify?
There's no fixed limit; it separates as many distinct voices as it detects. Accuracy is best with a handful of clearly distinct speakers.
Does it know people's names automatically?
No, it separates voices; you add the names. It may suggest a name that's mentioned in the conversation, which you confirm with one tap.
What if one person shows up as two speakers?
Merge the two labels into one. It's reversible.
Do the names appear in the summary and exports?
Yes. Once you name a speaker, the name is used in the transcript, the summary, action items and every export.
Does it work for Spanish and English conversations?
Yes. Both languages are detected automatically.
Know who said what
Upload a conversation and get a transcript by speaker, a summary and action items per person. Free to start.
Start free