Hey Nylas team,
I’m running into a pretty brutal issue with Notetaker: speaker attribution isn’t stable within the same meeting. Different transcript variants/segments disagree on who’s speaking at the same timestamps. That means any analytics built on “who said what” are unreliable.
This isn’t a theoretical nuisance. We have a feature ready to ship that depends on correct attribution, and this is the only blocker. If we can’t trust the labels, we’ve basically wasted weeks of work and will need to evaluate alternatives.
A few constraints:
-
Meetings were held on a mainstream video platform. Platform brand isn’t really the point—the issue is attribution drift within a single session.
-
Because this forum is public, I can’t share transcripts, or meeting artifacts here. If a staff member can DM me, I’ll provide the Notetaker IDs privately so you can pull exact examples on your side.
What I need from Nylas:
-
Is this a known issue? Please confirm.
-
Status/roadmap: Is a fix in progress, and what’s the expected path to stable attribution?
-
Any recommended workarounds (flags, params, or post-processing guidance) to keep speaker mapping consistent until a fix lands.
-
Decision signal: If there’s no near-term plan, please say so plainly so we can make an informed call (including switching providers if necessary).
Side note: I’ll be honest—needing to post potentially sensitive debugging details on a public forum to get support is… not ideal. I’ll keep things high-level here; happy to share specifics privately if a staff member reaches out.
Thanks in advance. This is a launch-blocker for us, so any clarity or workaround would be hugely appreciated.