recommendation for long running processes

[louis030195]

hi, i'm thinking in implementing pyannote-rs in

https://github.com/mediar-ai/screenpipe

which runs audio + transcription 24/7 over all audio devices

i'm still not super familiar in how it works, but ideally this would do:

• learn voice of each speaker over time (self supervised / unsupervised) and save the embeddings to disk
• every time some more audio chunk goes through the pipeline it updates the embeddings

does pyannote actually works like this? or it's just able to detect speakers like zero shot every time?

how much resource does it use? what about latency?