//
Reliability
Useful first drafts, built for review.
Earmark produces useful first drafts grounded in the meeting and the context you provide. Quality depends on audio, speakers, context, and how you define the task — so consequential work should be reviewed before it becomes the system of record.
This page explains how to tell what Earmark is capturing, where capture and generation can fall short, and how to review and correct what it produces.
At a glance
Grounded in the meeting
Every artifact is generated from your task, the transcript, meeting context, and your own personalization.
Capture you can see
A live transcript, floating widget, timers, and toasts tell you when Earmark is capturing — and when it isn’t.
Built to be reviewed
Treat each artifact as a first draft. Check decisions, owners, and dates against the transcript before sharing.
Easy to correct
Edit directly, fix attribution, or change the task prompt and regenerate from the full transcript.
Capture-state indicators
Before you rely on a transcript, confirm Earmark is capturing. These are the signals it shows today.
Indicator
Where
What it tells you
Live transcript
Meeting view
Text appears within moments of speech. If it stops updating while people talk, capture isn’t reaching the transcription service.
Live Widget
Desktop, floating pill
Capture is active and the sound bars move with audio. Toggle it in Settings → Capture.
Meeting timer
Meeting command bar
Elapsed time for the current session. Hover to see the 2-hour limit.
Mute microphone
Meeting command bar
Shows whether your mic is muted. While muted your voice isn’t transcribed; on desktop, system audio still is.
Stop Capture
Desktop tray or menu bar
Capture is running and can be stopped from outside the app.
Temporary meeting
Meeting header
This meeting won’t be saved to Earmark’s servers.
Reconnecting…
Toast
The transcription connection dropped and Earmark is retrying. Speech during the gap isn’t transcribed.
Time limit reached
Toast at 2 hours
Capture has stopped. Click Start Capture to continue.
Glance at the transcript in the first minute of every meeting. If you can see the other side of the conversation appearing, the rest of the meeting is very likely to capture the same way.
Supported audio scenarios
Scenario
Desktop (macOS, Windows)
Web
Remote call with headphones
Microphone and system audio
Your voice only
Remote call through speakers
Supported
Partial — remote voices only if your mic picks them up
In-person, one device in the room
Microphone
Microphone
Hybrid room with remote participants
Supported
Partial
Languages
English, or Multilingual (EN, ES, FR, DE, IT, PT)
Same
System audio capture is what lets the desktop app hear other participants when you wear headphones — browsers can’t capture system audio. The workspace transcription language is set by an admin in Settings → Workspace.
Speaker handling
Earmark doesn’t label speakers. On desktop, your microphone and system audio are mixed into one stream before transcription, so the transcript can’t tell voices apart. It breaks into turns at natural pauses, and uses attendee names, the meeting title, and context to attribute most statements.
Expect mistakes when:
Two people with similar roles discuss the same topic
People talk over each other
Someone speaks only briefly
When attribution matters — like owners of action items — confirm names out loud in the meeting and check them in the artifact.
Common failure modes
What goes wrong
Why
What to do
Other participants missing from the transcript
Using the web app with headphones, or the macOS Screen & System Audio Recording permission isn’t granted
Use the desktop app and grant the permission
Transcript stuck on “connecting”
A firewall, VPN, or TLS-inspecting proxy is blocking WebSockets to streaming.assemblyai.com
See Live transcription not working
Gap in the transcript
The network dropped; audio during reconnection isn’t buffered
Earmark retries automatically. If it gives up, resume capture.
Capture stopped mid-meeting
Sessions stop at 2 hours
Click Start Capture again
Capture stopped in one window
Earmark captures one meeting at a time
Keep one meeting capturing at a time
“We’re experiencing high demand”
The transcription service is temporarily rate-limited
Wait a moment and resume
Names, acronyms, or jargon misspelled
The speech model hasn’t seen your terms
Add attendees; set your role, bio, and company vision
Wrong words in non-English speech
Workspace is in English mode
Ask an admin to switch to Multilingual
Artifact misses or overstates something
The task was too broad, a moment fell in a gap, or the model inferred a conclusion
Make the task specific, pin key moments, and check the transcript before sharing
Grounding and context
Every artifact is generated from the same set of inputs:
Your task prompt — what you asked for, and in what format.
The meeting transcript — up to the moment of generation, or the full meeting once capture stops.
Meeting context — the title and the attendees.
Your context — your role and bio from Personalization, and your workspace’s company vision.
Pins — pinned moments, added as priority context.
Artifacts don’t currently include citations or timestamps that link statements back to the transcript. Use the transcript as the source of truth when you check facts.
Review workflow
Treat every artifact as a first draft. For consequential work — decisions, commitments, customer-facing messages, tickets — review it before it leaves Earmark.
1
Confirm capture
Skim the transcript for gaps, especially around reconnections or the 2-hour boundary.
2
Check the claims that matter
For each decision, owner, date, or number, find the matching moment in the transcript. Use the command menu (Cmd+K / Ctrl+K) to search.
3
Fix attribution
Correct anyone credited with the wrong statement or task.
4
Edit or regenerate
Make small fixes directly in the editor. If the artifact is off-target, change the task prompt and regenerate.
5
Share
Export or send it to your tools only after review.
Earmark doesn’t keep a version history for artifacts. Regenerating replaces the current content — copy any manual edits you want to keep before you regenerate.
Regeneration and editing
Action
How
Notes
Live updates
Turn on Sync live on a task during a recording
Regenerates about every 45 seconds as the transcript grows, and finalizes when capture stops.
Regenerate
Edit the task prompt
Produces a new version from the full transcript and replaces the current content.
Edit directly
Type in the artifact editor
Changes save automatically.
Focus on a moment
Click a pin
Pre-fills a prompt with that slice of the transcript.
Split by audience
Add more tasks
Create separate artifacts for engineers, leadership, or customers.
Troubleshooting
Live transcription not working
WebSocket and firewall issues
App appears disconnected or stale
Realtime sync blocked
Two-hour capture session limit
Long meetings
“Open Earmark” does nothing
Desktop sign-in handoff
Status and support
Earmark doesn’t publish a public status page or an uptime SLA yet. If capture or generation isn’t working:
1
Check the troubleshooting guides above.
2
Email support@tryearmark.com with the meeting time, platform (desktop or web, macOS or Windows), and any error message.
How we measure quality
We don’t publish a single accuracy percentage. Transcription and artifact quality vary with microphone quality, background noise, accents, crosstalk, vocabulary, language mode, and how specific the task is — a single number would hide those differences.
To evaluate Earmark for your team, run it on a representative set of your own meetings:
Capture — compare the transcript against a recording for a few minutes of each meeting, and note missed segments and misheard terms.
Artifacts — for each artifact, count the statements that are supported by the transcript, unsupported, or missing.
Mix of conditions — include remote and in-person meetings, headphones and speakers, and your typical languages.
Evaluate Earmark on your own meetings
Run it on a representative set of your meetings, then tell us what you find. We’ll help you structure an evaluation.