How to Transcribe Audio With Multiple Speakers and Label Who Said What
How to record well, transcribe and label who speaks using timestamps, and which tools identify speakers automatically (public data, Nov 3, 2026).
Sonorarium Team 6 min read
Short answer: Sonorarium transcribes what is said and puts a timestamp on every segment, but it doesn't separate speakers: it won't write "Maria:" or "Luis:" for you. The route that works with us is to record well (one microphone per person if you can), transcribe, and label who is speaking by hand, using the timestamps as your guide. If you need the labeling to be automatic, some tools do identify speakers; below are four, with their public data and the date we checked.
This works for a two-person interview, a five-person meeting or a focus group. The more voices and the more overlap, the more review work there is, and we say so up front so it isn't a surprise.
What Sonorarium does and doesn't do
- Does: transcribes the audio in the language you choose, with a timestamp on each segment. Click it and that exact moment plays, so finding a change of voice is quick.
- Does: lets you edit the text in the editor and download it as Word, TXT, SRT, VTT or JSON with your corrections (downloading and editing need your account). The Word file starts each paragraph with
[hh:mm:ss]; the TXT has no timestamps. - Doesn't: separate or name speakers. The AI summary doesn't either: it says what was discussed, not who said it.
1. Record with the end result in mind
Separating voices is mostly decided before you transcribe:
- One microphone per person if you can, even each person's phone (note: each file uses its own minutes). With only one, put it in the middle and within a meter of everyone.
- Have everyone say their name at the start. It takes ten seconds and gives you the key to the rest of the recording.
- One turn at a time. Overlapping speech is the hardest to transcribe and the hardest to attribute afterwards.
- Test for 30 seconds and listen back before you start. If the room echoes or is noisy, see how to transcribe noisy audio.
- Ask permission from everyone who can be heard before recording. What US and UK law say, and how to ask, is in is it legal to record a class, a meeting or a call?.
2. Transcribe the file
- Upload the recording to the meeting transcription tool or the interview transcription tool: audio or video, up to 5 hours and 2 GB.
- Choose the language that was spoken.
- Open the text: every segment has its timestamp.
The first 15 minutes are free when you create an account, with no card. If the file is longer, those 15 minutes are transcribed as a preview and the rest is unlocked from your account, paying only for the minutes still missing, while the audio is still stored (7 days).
3. Label who's speaking, using the timestamps
This is the method we recommend. With a recording of two or three people it's manageable; with eight, expect to go slowly. Do it within the 7 days the audio is kept: after that, the timestamps stay in the text but the segment no longer plays.
- Listen to the first minute and note what each voice sounds like (who introduced themselves first, who has the deeper voice, and so on).
- Move through the text by timestamp. A change of voice often coincides with a pause and a new sentence: click the segment's timestamp to hear it and decide.
- Write the name at the start of each turn, in one fixed format ("MARIA:", "LUIS:"). A single format lets you search and count later.
- Use tags for doubtful parts:
[overlap],[inaudible]or[Luis?]. It's better to flag a doubt than to misattribute a sentence. - Re-check the important attributions (quotes, figures, decisions) by listening to the segment again.
- Download when you're done. If you correct inside the editor, the Word, TXT and SRT files come out with your corrected text.
An example, fictional, of how a passage ends up. Before:
[00:00:04] Good morning, let's start.
[00:00:08] OK, here are the figures.
[00:00:13] Great, and the deadlines?
After labeling (made-up example, not a real transcript):
[00:00:04] MARIA: Good morning, let's start.
[00:00:08] LUIS: OK, here are the figures.
[00:00:13] MARIA: Great, and the deadlines?
For focus groups or research interviews there is a guide with conventions, tags and how to import into analysis software: transcription for qualitative research. For dissertation interviews, how to transcribe interviews for a dissertation or thesis. And for Teams, Zoom or Meet meetings, how to turn them into text.
4. If you need the labeling to be automatic
Some tools identify speakers on their own, which Sonorarium doesn't. This is data from their own pages, checked on November 3, 2026. It changes often: always check their site. We haven't tested them on your audio, and labeling quality depends on the recording.
| Tool | What its site says | Free or cost, per its site |
|---|---|---|
| Word (Microsoft 365) | Separates and labels speakers ("Speaker 1", "Speaker 2"…), which you can rename | With a Microsoft 365 subscription, up to 300 minutes of uploaded audio per month |
| Otter | Its Basic (free) plan lists "Speaker identification" | 300 minutes a month, up to 30 minutes per conversation and 3 lifetime file imports, on Basic |
| Notta | Its Free plan lists "Speaker identification" | 120 minutes a month, up to 3 minutes per conversation |
| noScribe | Distinguishes between speakers; uses pyannote | Free and open source (GPL-3.0); runs on your own computer, no cloud |
Two details: Word's feature is available in Word for Microsoft 365 on Windows (and, per its page, in Word for the web) and works in Edge and Chrome, and noScribe installs on your computer (Windows, macOS or Linux).
If your recording runs an hour or more, check the per-file cap before the minutes per month: on the free plans, Otter allows 30 minutes per conversation and Notta 3, so you'd have to split it. Sonorarium transcribes files of up to 5 hours in one go, with nothing to install and packs with no subscription: you pay only for the minutes you're missing. What it doesn't do is label speakers; if that matters most to you, those tools do something we don't. The free limits of each one are compared in free audio transcription limits, tool by tool.
What an hour-long meeting or interview costs
With a new account, the first 15 minutes are free when you create it, with no card, and 45 are missing. The 120-minute pack costs $6, and the 75 left over don't expire. All packs are on the pricing page.
FAQ
Will Sonorarium separate speakers in the future? We don't promise it. It doesn't today, which is why this article shows you how to label them yourself.
Does the AI summary say who spoke? No. It summarizes the topics but doesn't attribute sentences to people.
What if each person is recorded in their own file? It's the easiest way to know who is speaking: transcribe each file separately, and the file name tells you. Bear in mind that each file uses its own minutes: a one-hour meeting recorded on three phones is three hours of transcription. If cost matters, one good microphone in the middle of the table gives you a single recording.
What happens to the uploaded audio? It is deleted after 7 days; the transcript in your account stays until you delete it and isn't used to train models.
Tool sources: Microsoft Support (Transcribe your recordings), Otter, Notta and the noScribe repository on GitHub, checked on November 3, 2026.