Improved Speaker Diarization & Persistent Speaker Tagging

Multiple instances of other people being labeled as speaker zero even when I have no other speaker tags and a voice profile added. Reviewed audio files and something is chopping up the files and cutting speech parts off and that directly aligns with the changes back to speaker zero when it's not me.

At this point the way this is working it is no better than a personal note taking device where only I speak to it. My old compass AI device did that just fine. I want something I can build off of that will actually improve my life but I can't build off of broken transcripts and a device that can't recognize when I'm speaking vs someone else when my voice profile is the only one it has to go off of.

What is being said and who said it is the closest thing to truth the whole system has to go off of. This should be the foundation, it should be rock solid before anything else.

Started with not recognition own voice, then i let device keep running until battery runs out, turned on, “Introduced myself” again.

Now it can recognise my voice but not others. Even tho i have tagged the speaker before.

Please help

Request:

• Ability to reset or remove all speaker assignments for an entire transcript.

• Ability to select multiple messages and batch assign speakers in a single action.

Problem:

• If a speaker is incorrectly assigned, each message must be manually selected and reassigned one by one.

• If a speaker is incorrectly assigned, the “tag other segments” feature does not apply correctly, which makes correcting mistakes especially time consuming.

Related:

• Speaker selection currently defaults to alphabetical order. It would be helpful if it remembered the last used speaker or defaulted to “You.”

Problem / Context

When using Omi Desktop as my primary setup, the microphone input almost always represents me (the user). However, the system currently attempts speaker detection on microphone audio, which can result in unclear or incorrect speaker attribution in meeting notes.

In most real-world use cases:

• The microphone = the user

• All other voices come from system / speaker audio (calls, videos, meetings)

• There is typically only one person speaking on the mic

Because of this, meeting notes would be significantly clearer if microphone audio were consistently attributed to the user by default.

Proposed Solution

Add an option to label microphone audio as the user by default, bypassing speaker detection for mic input unless explicitly enabled.

• Microphone audio → Always attributed to the user’s name

• System / speaker audio → Continue using speaker detection as usual

• This assumes a single speaker on the mic (the user), which matches most desktop use cases

Optional Controls / Flexibility

To handle edge cases, add a toggle such as:

• “Assume mic audio is me (default)”

  • ON: Mic audio is always attributed to the user

  • OFF: Enable speaker detection on mic audio (for shared mic scenarios)

This allows advanced users to override the default behavior when:

• Multiple people share a microphone

• A different speaker is using the mic

• Group recordings are happening in the same physical space

Why This Matters

• Much clearer and more accurate meeting notes

• Reduces incorrect speaker labeling

• Matches how users actually use Omi Desktop in practice

• Simple default behavior with an escape hatch for edge cases

Summary

Assume one speaker on mic = the user, by default.

Let system audio handle everyone else.

Provide a toggle only when needed.

This small change would significantly improve transcript clarity and usability for desktop-first users.