MuniMins uses voice profile technology to automatically identify who is speaking in your meeting recordings. The system learns each speaker's voice over time, getting more accurate with every meeting you process.
How It Works
When you upload a recording, MuniMins runs voice identification — a process that separates the audio into segments by speaker.
Behind the scenes, MuniMins also extracts a voice "fingerprint" (embedding) for each speaker. This fingerprint is compared against any stored voice profiles for your board members and participants.
The "Speaker Training" feature will allow you to direct the system to learn to identify known speakers.
Importantly, only speakers that have enough voice data to be seperated will be shown. It is not an all-inclusive list of all the speakers. Remember, speaker training is to make future meetings more accurate, so if speakers are missing from this view it will not effect the minutes quality of that meeting.
Speaker Idenification
Before a voice profile is built the system will identify speakers based on context clues - for example if during a roll call the member identifies themselves the system will use that voice profile for the rest of the meeting.
Before voice profiles are build you will see each speaker that the system has identified with a tag for "HIGH" or "MEDIUM" confidence.
In this instance the system idenfied "George Washington" with high confidence and recognized he was a board member. You can verify this by playing a voice sample and then click "Confirm" to tell the system to remember how this speaker sounds.
When speakers are identifed via their voice the system will show a % match, rather than a confidence score.
In this meeting the voice profile for John Hancock matched with a 91% score.
The Training Period
On average it will take the system 2-4 meetings to build a good voice profile. Each meeting has slightly different audio, or, perhaps a member is remote, and overall if there is greater variety in the voice it will take longer to train.
| Meetings Confirmed | What Happens |
|---|---|
| 1st meeting | No profiles exist. You manually assign every speaker. Profiles are created. |
| 2–4 meetings | System suggests matches but with medium confidence. You confirm to strengthen profiles. |
| 5+ meetings | High-confidence matches are auto-confirmed during processing. No manual step needed. |
You can go to the "Members" page to view the status of each member's voice profile:
Tips for best results:
- Confirm speakers on every meeting for the first few meetings. This builds the strongest profiles.
- Use the dropdown, don't type names manually. Only dropdown selections link to voice profiles.
- Add board members and participants first (under Organization in the sidebar) so they appear in the dropdown.
How Profiles Improve Over Time
Each time you confirm a speaker, MuniMins stores their voice fingerprint for that meeting and recalculates a running average across all their confirmed meetings. More data points = a more distinctive profile = better automatic matching. As the system builds the voice fingerprint it will show fingerprint status in the speaker view.
The confidence system works on two factors:
- Voice similarity score — How closely the new audio matches the stored profile (threshold: 85% for auto-confirm).
- Number of confirmed meetings — More meetings build a more reliable profile (threshold: 5 meetings for auto-confirm).
Both thresholds must be met for fully automatic identification.
What About New Members?
When a new board member or participant joins, they start with no voice profile — just like when you first set up MuniMins. After they speak in a few meetings and you confirm their identity, their profile builds up and they'll be recognized automatically going forward.
If a member leaves and is replaced, simply add the new member under Organization → Members. Their voice profile builds independently.
Frequently Asked Questions
Do I need a microphone for each speaker?
No. MuniMins works with a single recording source — even a camera at the back of the room. The AI separates speakers by their voice characteristics, not by microphone.
What if the system identifies someone incorrectly?
Just correct it on the transcript view. Click the wrong name, select the right person from the dropdown, and confirm. This corrects the current minutes and improves the profile for future meetings.
Does speaker identification work with YouTube uploads?
Yes. YouTube and Vimeo link uploads are available on MuniMins, which includes voice identification. The audio is extracted from the video and processed the same way as file uploads.
Can I reset a voice profile?
You can delete the voice profile by removing the individual from members or participants, however, we do not recommend doing this. Every voice profile is being constantly updated, so going through the normal process of identifying the member at meetings will readjust the profile.
Speaker identification and voice profiles require the MuniMins plan.