Does Omi's transcript → action-items hold up on phone-call audio?

[baptistecristo]

Thanks for open-sourcing Omi. I'm building a small open-source assistant you reach by calling an ordinary phone number: no app, no wearable, bilingual EN/FR, EU-hosted. The part I want is the one you already do well, turn a conversation into a summary, pull the action items, let someone ask about it later.

Before I dig into the code, one question from your experience. Does that pipeline hold up on short, phone-quality audio, where both sides are often mixed into one low-fidelity channel, unlike a wearable's clean mic? That's the piece I'm unsure carries over. I'm not asking anyone to build it, just whether you've hit this already and where in the repo to start. Thanks.

[baptistecristo]

Also, the repo issue's link is baptistecristo/AI-agents-for-dumbphones#29 is you want to give a hand :)