identify persons in audio

[louis030195]

No description provided.

[linear]

MED-92 identify persons in audio

[NicodemPL]

I tried myself with paynnote audio - this works pretty good. fully local. but python based.
https://github.com/pyannote/pyannote-audio

highly recommended - tried different systems (also paid) and this one is really efficient and delivers good quality.

On top of this - once you have separate audio - microphone and display this is even more promising for better quality meeting notes.
I've developed a Python tool that combines Whisper transcription and Pyannote diarization to create comprehensive meeting transcript. This automated system transcribes audio, identifies speakers, and integrates the results, laying the groundwork for AI-assisted prompt for good notes generations.
Still got some issues on my side but its basically working and 100% local. So this is doable for sure.
And it beats Rewind.ai / Limitless for sure :) Locally.

[louis030195]

/bounty 100

definition of done:

• screenpipe-audio has some code that identify speakers - i guess after the transcription?
• this is sent to the screenpipe-server which would insert speakers into DB
• which is then returned in db queries & api

rules:

• use rust, local
• do not use too much compute, screenpipe must still be usable on normal consumer hardware (eventually if it's possible to use GPU/NPU...)
• ideally separate file in screenpipe-audio
• works on all OSes

[algora-pbc]

💎 $100 bounty • Screenpi.pe

Steps to solve:

1. Start working: Comment /attempt #306 with your implementation plan
2. Submit work: Create a pull request including /claim #306 in the PR body to claim the bounty
3. Receive payment: 100% of the bounty is received 2-5 days post-reward. Make sure you are eligible for payouts

Thank you for contributing to mediar-ai/screenpipe!

Add a bounty • Share on socials

[kernel-loophole]

@louis030195 is this issue is still open ,would love to work on this

[louis030195]

@EzraEllette is on it i believe

[EzraEllette]

Doing this now. Almost ready for a PR. Writing Tests.

[EzraEllette]

So it seems like all of the hallucinations match to the same speaker, which could be useful for determining if part of a transcript is hallucination... I will update my branch, but I need to do performance improvements because right now I am segmenting speech then performing stt on each segment. @louis030195