Guys, I am having a ton of trouble with speaker tagging. The whole changing speaker tags across conversations is a real pain, so much so that I almost done want to do it. If I delete too many speakers I need to go back and tag individual segments. If I delete all speakers I can’t add new ones to retap. Is there a guide somewhere on how to do this well???
btw I love the product, this is just a real rock in my shoe.
hey @jpalusci so we are pushing a change which create a new speaker for a specific conversation (if its being renamed and associated with another meeting) just currently under review!
Hey, just keeping things clear about how the speaker management works.
When you rename a speaker, we try to keep things simple and avoid messing up your older conversations. Here’s how it works:
If it’s a speaker you made:
We just change the name everywhere. Easy.
If it’s a speaker the system detected and it only shows up in this one recording:
We rename it normally.
If the system detected that same speaker in multiple recordings:
We only rename it for this recording so your older conversations don’t suddenly change.
(We make a new speaker with your new name and use it here.)
We clean up any leftover unused speakers so things don’t get messy.
Thanks Chay. It sounds simple but that’s not the outcomes I’m experience when I am adjusting transcripts. Hopefully the upcoming update will address this and I’ll check back in to see if it’s gotten any better. Would be helpful if there was a Tips and Tricks on how to best manage speaker names that considers all the functionality around this
@chay please look into this
First let me say I love the device, I see the vision, and I believe in the product and team behind it.
With that being said, I’ve more less given up trying to label people for now. I’ve never done it. My Pocket always calls people speaker 0, speaker 1, speaker 2, etc. when it’s me, my wife, and my child… At this point, after hours and hours of recording, it should be labeling us all easily, no?
I only just gained access to an intuitive place to adjust speaker tags in settings in this last update. I took a peek in and it’s a mess in there. I’m listed many times, there’s no memories attached to any names, there’s tons of random names again with no memories. I still can’t figure out how to change a tag. I can’t read what it says when I click a name easily. I had to screenshot and zoom in; if my memory serves, it read in broken English roughly something along the lines of, “keep recording and Pocket will tag automatically.”
it would be really nice if I could just go into a memory, [tap] “speaker 0” and [select] who it is from a drop down of common contacts that I could have preloaded myself or the AI could suggest.
I’m having similar issues. I’ve got the same name listed multiple times, just in different colors. Is there a way to merge speakers?
This is available in individual transcripts, but for some reason it’s not available in speaker management. Which is really where I need it. I have 2+ copies of each speaker, likely the backend recognizing them as different people and me relabeling them with their name.
I agree completely. I would personally just want the team to pause other feature requests for week to get this sorted out.
There a lot of great things going on with Pocket and the app, but speakers are frankly a mess atm
I see what you mean. I have 6x of myself and multiple other speakers that are all duplicated.
I also see a bunch of random Speakers, with 0 summaries attached to them. There’s about 50 of those
Wanted to share some more specific feedback on the speaker labeling architecture and suggest a few improvements to fix the current data fragmentation. @mhirrrr @chay
Deprecate “Local” Management & Centralize Workflow
The per-memory speaker view is currently confusing and seems to allow for the creation of isolated, temporary speaker labels that don’t persist or inform the model.
Enforce a Single Source of Truth (SSOT)
If I label a speaker in a memory, it must immediately link to or create a Global Profile. Currently, the AI might label “Matthew” as “Speaker X” in one recording and “Speaker Y” in another because the system treats them as separate profiles.
Fixing the “Delete” & Assignment Logic
Because of the current local/global split, users are trying to “delete” speakers to clean up the mess, which breaks the segment data by leaving it orphaned.
Yup. See post above. I think this would solve our problems, hopeful they’ll give it some attention and love
Yup, this seems to make much more sense and would reduce the load and storage
Does this mean “everywhere” as in the entire library of recordings, or just within that transcription/recording?
I agree 
! It’s a real obstacle to an otherwise amazing product. I reported similar issue before and agree this could use some additional attention.
We will look into this with utmost priority
That’s my issue as well. I have a bunch of conversations I didn’t carefully go through as it got to be too much. Now, the idea of going back through them all makes my head hurt.
This is very well thought out.
This is where I’m at now, too. Pocket would be really cool if it worked, but right now, it’s just a novelty since it doesn’t work. I’ve just shelved it and I’ll check back later. I just don’t have the time to deal with the frustrations of trying to figure out how to make it work in a useful way.
