Know who's speaking.

From “Them-1” to real names

Every recording starts the same way: echo99 keeps your microphone and the other side of the call as separate tracks, labels your mic “Me”, and splits the other voices into “Them-1”, “Them-2”, and so on. Speaker identification is the layer on top: it turns those anonymous “Them-N” labels into the actual names of people you’ve met before.

How recognition works on your Mac

Put a name to a voice once, and echo99 stores a compact voiceprint for that person in your local gallery. On every later recording it compares each speaker against that gallery and, when a voice is a confident match, labels the transcript with the person’s name automatically. There is no cloud service in the loop. The matching runs entirely on the Mac in front of you.

When echo99 is not sure, it deliberately does nothing: the speaker stays “Them-N” rather than risk putting the wrong name on the wrong words. Identity only ever adds a name it’s confident about.

How to name speakers

1. Keep Recognize speakers across recordings ticked when you set up echo99 (it is on by default for new installs), or turn it on later in Config → Transcription.
2. After a call has been transcribed, open it in the Recordings window. In the Speakers section, play a sample of each far-end speaker and type their name. The transcript is relabeled right away, and the person joins your gallery.
3. On later calls, echo99 matches that voice and fills in the name itself.
4. To add someone from a recording you made earlier, open Manage Voices… and choose Add from recording….

Recognition uses the standard Speaker diarization model, so it needs that model installed even if you picked a different speaker model for transcription. If it is missing, Config → Transcription offers to install it.

You stay in control

• Manage Voices: rename or remove anyone in your gallery from the Voices window, opened with Manage Voices… in the menu-bar panel or in Config → Transcription. A rename also updates their name in past transcripts.
• No names without you: a voice becomes a named person only because you named it; anyone you haven’t named stays “Them-N”. If you don’t want voiceprints at all, untick Recognize speakers across recordings during setup or turn it off in Config → Transcription.
• Corrections are easy: relabel a speaker whenever a match isn’t what you expected; your correction is the source of truth.

Private by design

Voiceprints are biometric data, and echo99 treats them that way: they are stored locally on your Mac, never uploaded, and never used to train anything. They leave your Mac only if you export them yourself (Manage Voices → Export Voices…). Each voice you name also keeps a short sample of that person speaking, taken from the call, so you can listen to it in Manage Voices; samples are never exported. Everything else about speaker identification happens locally, like recording and transcription, so it works with Wi-Fi off and nothing is sent to anyone else’s server on its own.

To remove someone completely, delete them in Manage Voices and choose Delete and remove from past recordings, which also relabels their lines back to “Them-N” in every recording in your current recordings folder. Delete all speaker recognition data… in Config → Transcription clears the whole gallery and removes speaker data from every recording in your current recordings folder, relabeling those speakers back to “Them-N”.

Limits

• Voice matching is strong but not infallible: very similar voices, or poor audio, may stay “Them-N” until you name them.
• Recognition improves with a clean sample: the clearer the recording you first name someone from, the better later matches will be.
• It identifies the people you have named; it does not attempt to guess the identity of strangers.

New to echo99? See the getting started guide.

Related guides: Local transcription, Transcribe interviews, Meeting notes without a bot