Meeting Intelligence
Dual-channel speaker diarization. Your mic and their audio on separate channels. Live transcription with speaker labels. No bot joins. No audio leaves your Mac.
Dual-channel diarization
Your microphone captures your voice. System audio captures everyone else. These two channels are transcribed separately, then merged with NVIDIA Sortformer diarization — achieving ~11% Diarization Error Rate without any cloud processing.
The dual-channel approach means Resonant already knows which audio is you before diarization even runs. This dramatically reduces attribution errors compared to single-channel solutions.
I think we should move the JWT migration up to this sprint. The old tokens are causing flaky test failures.
Agreed. How long do you think the migration will take?
Two days if we keep the old validation as a fallback. I’ll have a PR by Thursday.
Can we run both validators in parallel during the rollout? That way we don’t break existing sessions.
Yeah, I was thinking a feature flag. Old path stays live until we’re confident.
How it works
Resonant detects when you join a call in Zoom, Meet, Teams, or any meeting app. Recording starts automatically — or manually with a hotkey.
Your mic audio and the system audio (remote participants) are captured on separate channels. This separation is what makes speaker attribution possible.
NVIDIA Sortformer identifies who is speaking in real time. Combined with the dual-channel split, you get speaker labels with ~11% DER — without any cloud processing.
Full transcript with speaker names, timestamps, and topic segmentation. Saved locally, searchable in your workspace, queryable via MCP.
What makes it different
Cloud transcription services add a bot participant to your meeting. Everyone sees it. Everyone knows. Resonant records from your Mac's audio channels — invisible to other participants.
The recording indicator hides automatically when you share your screen. No one sees a "Recording..." badge. The transcription continues silently in the background.
A live notepad appears during recording. Type thoughts, action items, or questions while the meeting is transcribed. Notes are saved alongside the transcript.
Text-based echo suppression detects when the same words appear on both channels (your mic picking up speaker output) and deduplicates them automatically.
Every meeting is full-text searchable in your workspace. Find "that thing Sarah said about the API" across months of meetings in seconds.
Your AI assistant can pull meeting transcripts directly via MCP. "What did we decide about the deadline?" gets a real answer with the exact quote.
Comparison
Privacy
Cloud meeting transcription services receive your meeting audio on their servers. Every word — confidential discussions, personnel conversations, strategy sessions — passes through infrastructure you don't control.
Resonant processes meeting audio entirely on your Mac. The transcript is stored locally. You decide what to keep, what to search, and what to share. No vendor has access to your meeting content.
For sensitive environments: HIPAA mode encrypts all stored transcripts, suppresses telemetry, and provides a full audit log. Compliant by architecture, not by policy.
Free. Local. No bot.
Dual-channel diarization with speaker labels. Searchable. MCP-queryable. All on your Mac.
Requires macOS 14+ · Apple Silicon