All-Ears is a set of composable tools to transcribe and summarise meetings. - A daemon that records audio sources - `transcribe` to transcribe the audio (with diarisation) - `cleanup` to clean the transcription, using known word dictionary - `summarize` to summarise the transcript
You can switch out models for transcription or diarization. You can use whatever prompt you want for cleanup and summarization.
One feature I'm still working on: There's a browser extension. It automatically starts and stops a meeting transcription when you join a Google meet/zoom/teams in the browser. It extracts the audio streams from each participant, so the transcription engine gets audio for each participant rather than the combined audio. Also it automatically gets their name out of the DOM so the transcript has names assigned automatically.
I'd also like a MacOS menu bar icon, and it's built to allow any front-end to hook in, but that's for later.
If you like the idea, stars and PRs are welcome.