Augmem v3 Enroll Speakers

Supervised Import

Enroll Speakers From Labeled Audio

Upload one recording plus a timestamped labeled transcript. Augmem will register the source, persist named transcript segments, optionally extract voiceprint embeddings from those exact spans, and rebuild semantic chunks from resolved transcript text.

Upload

Import Labeled Session

Transcript Format

Accepted Shapes

Use JSON or CSV with speaker-attributed segments and explicit times.

JSON

[
  {"speaker":"Alice","start":0.0,"end":4.2,"text":"Opening line"},
  {"speaker":"Bob","start":4.2,"end":8.9,"text":"Reply"}
]

CSV

speaker,start,end,text
Alice,0.0,4.2,Opening line
Bob,4.2,8.9,Reply

Outcome

What This Writes

Registers a manual source in SQLite.

Stores transcript rows and resolved speaker segments.

Creates speaker turns from the labeled time spans.

Optionally stores voiceprint embeddings for each qualified segment.

Optionally builds semantic chunks from the resolved transcript.