Managing personas
Every conversation is split between two personas — Agent and Customer. Enrolling voice samples for a named agent lets the pipeline recognise them and name them in the transcript instead of a numbered speaker; personas belong to one voice application only.
- The Voice IntELIEgence module visible in your left-hand navigation.
- A voice application, and a short, clean recording of each agent — WAV, MP3 or M4A, up to 25 MB per file.
If Voice IntELIEgence isn't in your navigation, your role doesn't have visibility for it — ask a workspace administrator, or see Permissions and module visibility.
Steps
Open Settings, then Personas, and turn on Enable speaker identification. Select Save changes — a dialog then offers Enroll agents, which opens the roster.
The roster is its own item below Settings: select Personas. Two fixed role rows sit at the top for Agent and Customer; Enrolled voices below lists each person’s Role, Samples, Status and Last used.
Select Enroll agent, enter the person’s Name, add up to three files under Enrollment sample, then select Enroll. Every enrolment is an agent — there is no customer roster.
One sample leaves the row Pending; two or more make it Enrolled. It reads Processing while the voiceprint builds. Add sample in the row menu tops up a Pending voice; Delete voice removes one.
What success looks like
- The row reads Enrolled in Enrolled voices, with Samples at two or more.
- New conversations name that person against the agent speaker.
Where a persona is referenced
In a conversation’s Speakers panel, the Agent dropdown lists your enrolled voices; choosing one asks Reassign speaker?, warns that it re-runs part of the pipeline, then needs Reassign & Re-run. It is disabled unless the transcript has an agent speaker, at least one voice is Enrolled, the conversation is neither processing nor approved, and the application runs the speaker-identification stage — the Pipeline on Settings → Application details shows Speaker map as Skipped when it does not. The Customer row is a label, not a picker. A conversation is pinned to the application version it started under, so enrolling a voice today never relabels one already processed — reassigning by hand is the only route.
If something goes wrong
| Symptom | Likely cause | What to do |
|---|---|---|
| Speakers stay unnamed on new conversations. | Enable speaker identification is off, or nobody is Enrolled. | Turn it on under Settings → Personas, then enrol two samples per agent. |
| A row sits at Pending. | It has one sample; two are needed. | Select Add sample in the row menu. |
| A row reads Failed or Not enrolled. | The samples could not be turned into a voiceprint. | Delete voice, then enrol again from a cleaner recording. |
| An older conversation names the wrong speaker. | It ran under an earlier version of the application. | Reassign it from that conversation’s Speakers panel. |
Next
- Voice application settings — where Enable speaker identification lives.
- Browsing conversations — where the names appear.