Description
Taking notes in a meeting means not really listening, replaying the recording afterwards eats time, and note-taking bots have to join the call — awkward for everyone, and the audio ends up in someone else's cloud. Earshot takes another approach: it listens to one app's audio on your Mac, so whatever Teams, Zoom, Chrome or Discord plays becomes text, with no bot in the call and no audio leaving the machine.
When the call ends it tells the voices apart, picks up real names in Teams, and has Apple's on-device model write a summary. If you want it live, open the subtitle window and lines appear a moment after they are spoken.
One click: The menu bar lists apps playing sound right now; click one to start.
After the call or live: By default only audio is saved during the call so the Mac stays cool, and the transcript is written once it ends; live mode is a switch away.
Speakers told apart: Diarization runs on the Neural Engine after the call; Teams names come from its live captions, others become "Speaker 1", "Speaker 2" and can be renamed.
Automatic summary: Apple Intelligence's on-device model writes a short summary in the language of the call.
Only that app: It taps the app's own audio output, so notifications, music and your mic stay out unless you choose to include your microphone.
Live subtitles: A small translucent window shows what is being said, using Apple's speech recognizer for speed or a Kyutai model for quality.
Filed by calendar: Sessions are named after their calendar event, with TXT, Markdown and SRT export; drop in audio or video files to transcribe them too.
Crash-safe: Audio is written as it arrives, so a crash loses only the last fraction of a second and the session resumes on next launch.
Requirements: The Mac app needs macOS 26 on Apple silicon, and there is a Windows build; transcription uses whisper.cpp with a separately downloaded Whisper model. It is not notarized, so the first launch needs approval in System Settings.
When the call ends it tells the voices apart, picks up real names in Teams, and has Apple's on-device model write a summary. If you want it live, open the subtitle window and lines appear a moment after they are spoken.
Features
One click: The menu bar lists apps playing sound right now; click one to start.
After the call or live: By default only audio is saved during the call so the Mac stays cool, and the transcript is written once it ends; live mode is a switch away.
Speakers told apart: Diarization runs on the Neural Engine after the call; Teams names come from its live captions, others become "Speaker 1", "Speaker 2" and can be renamed.
Automatic summary: Apple Intelligence's on-device model writes a short summary in the language of the call.
Only that app: It taps the app's own audio output, so notifications, music and your mic stay out unless you choose to include your microphone.
Live subtitles: A small translucent window shows what is being said, using Apple's speech recognizer for speed or a Kyutai model for quality.
Filed by calendar: Sessions are named after their calendar event, with TXT, Markdown and SRT export; drop in audio or video files to transcribe them too.
Crash-safe: Audio is written as it arrives, so a crash loses only the last fraction of a second and the session resumes on next launch.
Requirements: The Mac app needs macOS 26 on Apple silicon, and there is a Windows build; transcription uses whisper.cpp with a separately downloaded Whisper model. It is not notarized, so the first launch needs approval in System Settings.
