Speaker identification stopped working?

I just synced a very long meeting to my phone, and all of a sudden the pocket is making no attempt to identify speakers/assign speaker numbers. Each section of transcript is just “tag speaker”. Over 360 different segments, over a 3 hour meeting.

There is no way I can manually add speaker tags to 360 segments, let alone remember who said every line in a 3 hour meeting.

Was this something that changed with the update? This was never an issue before- when speakers would be labeled 0-5 and tagging one could tag all of the same speaker through the whole meeting. This change dramatically reduces the utility of pocket for meetings.

1 Like

I would email hey@heypocket.com

Wait, it is supposed to identify speakers automatically? I thought you had to do it manually…

hey @DaveNgl if we have a match from previous recordings and names are tagged we automatically tag them

Does updating segments speaker label improve the AI match detection?

Regarding auto tagging- That is very hit or miss for me (and not all that accurate… though I suspect it will get better with more data).

what I am referencing in this post is that instead of it splitting out speakers 0-3 and then tagging all of the segments with the speaker number- it just says “tag speaker”. In the previous speaker system, once you named speaker zero, it would rename all the speaker zeros in the conversation with one click. Now it’s an individual click (and having to remember who said what) for every segment… and which in my case was 300+ unique segments

1 Like

Yes I noticed this too. This is a big regression

Hey Guys so Speaker Identification becomes very inaccurate for meetings over 3 hours due to which we do not do the same that is what seems to be happening f here

This isn’t what @Gus is referring to. Pocket used push a change in speaker label from “Speaker 0“ to “dps“ to all segments that were originally labeled “Speaker 0“ in one motion. Now it seems to require manually updating EVERY segment individually.

hey @dps just checked that we do do that can just mail me an video if thats not whats happening

from my convo with Gus in the chat seems like his recording was over 3 hours

I’ve never had a match at all. I emailed support so they can look at my latest recording.

The notes have never identified speakers for me … only the transcript. This is going to be a huge problem with chat, because it seems to think all speakers are me.

Yeah, the transcripts don’t even identify me. I’ve even regenerated my own voice print…

1 Like

Hey @DaveNgl just a voiceprint isn’t enough for identifying speakers we always suggest tagging speakers across multiple conversations so there is a voice profile for a user which we can use for matching!

Do you have a suggestion for when multiple speakers are included in a section that only has one speaker tag? That’s the case for the majority of my transcripts, so I’m not sure how useful tagging one person will be if half of the audio is one voice, and half is another voice (or often more than two speakers).

We will be pushing something to split segments over the Monday so that will be helpful!

1 Like

Excellent!

You guys rock!

Same. I am never identified

1 Like

@mhirrrr I just read this and realize this is an interesting application.

So to ensure I understand - if I have multiple conversations with the same people I should always tag them the same? Like Joe should always be labled Joe, not Joe or JD or Joseph interchangeably?

The reason I ask - in some conversations I want to know that it’s Joe, but in other conversations for ease of sharing I may only use the initials JD ( first and last name initials ) to de-identify Joe in the summary.

For optimal long-term identification of Joe’s voice is there a better way to do what Im doing or does having one persons voice with multiple IDs/tags cause wrinkles, or does/will Joe’s three IDs eventually all match back to “Joe” through repetition?

1 Like