Live transcripts of your online calls, made on your Mac.
Voice writes down Zoom, Meet, Teams and Discord calls as they happen, with you and the other side kept apart. Your agents can read along. It does fast dictation too.
I wanted really fast transcription on my own Whisper pipeline, one I could hack on. Handy and OpenSuperWhisper are my favorites and I learned a lot from them, but I wanted something simpler. With Blend, Voice also hears both sides of an online call, so it writes down you and them separately.
Online calls, both sides apart
With a Blend split mix, you're Me and everyone else is Call. Voice asks when an online call starts, or press Option+Shift+`.
Your agents read along
Claude Code and other agents follow an online call live and search past ones over MCP.
No key throws a recording away
The audio is on disk as you speak. After a crash or a quit, the next launch transcribes it.
Dictation, and how it works
Press Option+` in any text field, talk, and press it again. Whisper large v3 turbo runs on your Mac with your key terms in its prompt, and Voice decodes the segments it's unsure of a second time and shows you what each retry changed.
Rough edges: Apple silicon only. Both sides of an online call need Blend. The speech model is a 1.6 GB download.
Is the code public?
Yes. The code is source available under the Functional Source License, so you can check what Dictate does with your audio. You get the signed build, which updates itself, by email.
Your audio never leaves your Mac, and there are no accounts. Voice goes online only to download the speech model and to check for updates. The update server counts update checks by app, version and country and never stores your IP address.