After a meeting, sorting out "who said what" always gets pushed back. The AI Meeting Notes tool takes a single recording and gives you a transcript, speaker labels and a draft of meeting minutes in one go. This guide covers how to record well, how to get minutes out of the tool, and a template for the final notes.
1. How to record a meeting well
- Tell participants beforehand and get their consent. Recording laws differ by country and, in the US, by state: some places only require one participant's consent, while others require everyone's. The simple rule is to tell everyone you are recording and get their OK before you start.
- Put the mic (or phone) in the middle of the group, not too far away. The farther it is, the worse the audio and the harder speaker separation becomes.
- Have people speak one at a time. Speaker labels get much more accurate. When people talk over each other, neither AI nor humans can tell voices apart easily.
- Pick a spot with little constant noise like AC units, fans or projector fans. If there is a lot of noise, try the Voice Noise Remover after recording.
- On a phone, check storage and battery and turn on Do Not Disturb so notification sounds do not end up in the recording.
- If the meeting will run well over an hour, consider recording in parts. Very long files take a long time to process.
2. Turn the recording into draft minutes
- Add a file to AI Meeting Notes or record right there with your mic.
- On a PC, choose the larger model (small) and set the language to match the meeting. Turn on speaker labels; if you know the number of attendees, set the number of speakers for better accuracy.
- When it finishes, read the conversation grouped by speaker, press ▶ to listen and fix wrong sentences. Rename speakers to real names or roles.
- If more speakers were detected than were really there, use "Merge speakers" in the results.
- Skim the automatic summary: key sentences, keywords, to-do candidates and decision candidates. Delete or fix anything that does not match the meeting. The extraction is rule-based (word frequency and sentence patterns), not AI writing, so it is not perfect.
- Save as meeting notes in Markdown for a ready-to-edit template. You can also save TXT or SRT (with timestamps).
3. Meeting minutes template
| Section | What to write |
|---|---|
| Date, place, attendees | Meeting date, place and the attendee names you confirmed from speaker labels |
| Agenda | Topics discussed in this meeting |
| Key discussion | A summary per agenda item, using the extracted key sentences |
| Decisions | Check the decision candidates and keep only what was actually agreed |
| Action items | Use the to-do candidates and fill in owners and deadlines yourself |
| Next meeting | Date and what to prepare |
The Markdown export lists date, attendees and length, key sentences, keywords, to-do candidates, decision candidates and the full conversation by speaker in that order, ready for editing.
4. When speaker labels are wrong
Speaker separation is AI that splits segments by voice characteristics, so it struggles with overlapping speech, similar-sounding voices and low-quality phone or video-call audio. If one person is split into two speakers, merge them. If two people ended up in one speaker block, move the sentences by cutting and pasting the text. Reassigning individual sentences to a different speaker is not supported yet.
5. Storing them
If the meeting includes personal data or confidential information, keep the recording and notes somewhere secure and delete them when they are no longer needed. FreeSoundTools does not store your files on a server, but downloaded recordings and notes stay on your device.
Transcription, speaker labels and the summary are all automatic. For contracts, disciplinary matters or disputes, always check the original recording.
Sources and date checked
Checked: October 5, 2026
- OpenAI Whisper — https://github.com/openai/whisper
- pyannote-audio (segmentation-3.0) — https://github.com/pyannote/pyannote-audio
This article is general information, not legal advice.
Related guides: how to convert a recording to text · meeting & lecture recording workflow · speech-to-text accuracy
Frequently asked questions
How does it know who is speaking?
An AI model (pyannote segmentation) splits the audio by voice characteristics. It works best when people speak one at a time and the mic is not far away.
What if it finds too many speakers?
Use "Merge speakers" in the results to combine them, and rename speakers to real names.
Does an AI write the summary?
No. Key sentences, to-dos and decisions are picked out by rules based on word frequency and sentence patterns, so review them before using.
Is my meeting uploaded anywhere?
No. Everything runs in your browser; files are not sent to a server.