You just transcribed your meeting. You open the file and face a wall of text. Sentences flowing one after another with no indication of who’s speaking. “We could do this differently” - but who said that? Your manager or the intern? “I’ll handle it” - great, but who exactly? You end up re-reading everything trying to mentally reconstruct who said what. Might as well listen to the recording again.

That’s the problem with most transcription tools. They convert audio to text, period. The result is technically correct but practically useless as soon as more than one person is talking.

The difference between transcribing and understanding

Transcribing means turning sound into words. Any service can do that today. But understanding a conversation is something else entirely. It’s knowing that this sentence came from Marie, the response came from Thomas, and the final decision was made by Julie. Without this information, you don’t have meeting notes, you have word soup.

For meetings, this is critical. Who made which commitment? Who raised that objection? Who needs to follow up with whom? These questions are impossible to answer if your transcript doesn’t distinguish between participants. You go from a tool that saves you time to one that wastes it.

Why so few services offer it

Speaker detection, what’s called diarization, is technically much more complex than basic transcription. It’s not enough to recognize words. You have to analyze each person’s vocal characteristics, distinguish them from one another, and maintain that distinction throughout the conversation.

This requires more advanced models, more computing power, and more processing time. Result: it’s more expensive to run. Many services prefer to offer basic, fast transcription rather than intelligent transcription that takes a few extra seconds.

Cosmonote’s choice

We made a different choice. We want you to leave a meeting with real notes, not a text file to decipher. So yes, our models are slower. Yes, it costs us more. But the result is there: each contribution is attributed to its author, exchanges are readable, and you immediately know who said what.

This is particularly useful when you use Ask AI afterwards. You can ask “what did Marie propose?” or “what points did the technical team raise?”. Without speaker identification, these questions would be meaningless.

What it actually changes

With standard transcription, you have to re-read the entire document to understand the flow of conversation. With speaker detection, you can quickly scan, spot the contributions from the person you’re interested in, and get straight to the point.

For action items, it’s even more obvious. “I’ll handle it by Friday” means nothing if you don’t know who said it. With identification, you know exactly who to follow up with if it’s not done.

And for meetings you couldn’t attend? You can read the notes like an actual conversation, not like a badly formatted TV script. You understand the dynamics, the friction points, the consensus. You catch up in minutes instead of having to call a colleague for a summary.