Transcription quality depends on the audio Clarify receives. These tips apply to both the bot recorder and the desktop recorder.
What affects accuracy most
Background noise is the biggest factor in word accuracy. Static, fans, keyboard clicking, echoing rooms, and ambient conversation all reduce how accurately speech is transcribed.
Speaker labels are a separate matter. They come from the meeting platform's participant information, not from the audio, so improving audio quality raises word accuracy without changing who a line is attributed to.
Overlapping speech loses words and attribution. When two people talk simultaneously, some words drop and the overlapping speech may be attributed to a single participant.
Low-volume participants - dial-in callers, people on poor connections, or quiet speakers - produce less reliable transcripts. Transcription works from the audio mix the meeting platform provides.
Bot recorder tips
The bot receives audio from the meeting platform's mix - the same mix all participants hear.
Ask participants to use headphones. Speakers create an audio feedback loop that degrades the entire meeting mix. Headphones eliminate this.
Mute when not speaking, especially in noisy environments. Ambient noise from multiple open mics compounds.
Bluetooth headsets can introduce compression artifacts. USB or wired headsets produce cleaner audio.
Phone dial-in participants typically sound worse in the transcript - their audio is compressed more aggressively than VoIP.
Room acoustics matter for in-person meetings. Hard surfaces (glass, concrete) cause reverb. A carpeted room with soft furnishings produces a cleaner recording.
Desktop recorder tips
The desktop recorder captures audio directly from your machine - what you hear combined with your microphone.
Use "Join and record" instead of "Quick Record" for scheduled meetings. Both live in the Clarify desktop app (menu bar on Mac, system tray on Windows). "Join and record" waits for the meeting window to be detected, improving both video capture and audio setup.
Keep the participant panel visible. The app identifies speakers by reading display names on screen. If the participant panel is hidden or minimized, some speaker labels may be missing.
Keep the meeting window in the foreground while recording. Switching away doesn't stop the recording, but window-detection for speaker names relies on the meeting UI being on screen.
Conference rooms and in-person meetings
When several people share a room and join from one device, the meeting platform reports a single participant. Clarify labels speakers per participant, not per voice, so every line of the transcript carries that one name. There is nothing for speaker separation to work with.
This applies to both recorders. The bot reads participant names from the meeting platform; the desktop app reads display names shown on screen. In a shared room, both see one name.
Audio quality has the same root. One microphone covering a room at a distance captures quieter, more reverberant speech than a headset at close range, which lowers word accuracy on top of the missing attribution.
Have each person join from their own laptop with a headset, muted unless speaking. Everyone is then a separate participant with their own name, and speaker separation and audio quality both come out close to a normal remote call.
If the room has to share one device, the conversation is still captured, transcribed and summarized. Only the speaker labels are affected: everyone in the room shares the one name, while anyone dialing in from their own device is still labeled separately.
Joining from a laptop in companion mode does not add speaker separation. Google turns the microphone and meeting audio off on a companion device to avoid echo, so no speech is attributed to it, and the desktop app on that laptop receives no meeting audio to capture. The room still joins as one participant, so the transcript is unchanged.
Language detection
Both recorders detect the spoken language automatically - no configuration needed.
Bot recorder: Multi-language transcription on Growth and above. On other plans the bot transcribes in English only.
Desktop recorder: Supports English, Spanish, French, German, Italian, and Portuguese on all plans, on both Mac and Windows.
Summaries and action items are generated in the language of the transcript.
See Multi-language meeting support for full details and plan requirements.
Limitations
Shared rooms produce one speaker label. When a room joins as a single participant, all speech is attributed to that one name. See Conference rooms and in-person meetings above.
Transcripts can't be manually corrected or re-edited in Clarify. Accuracy reflects the audio quality at record time.
Existing recordings can't be re-transcribed with different settings. Summaries can be regenerated (see Meeting summary templates), but the underlying transcript is fixed.
Speaker labels depend on display names in the meeting platform. If a participant used an alias, their label will show what the platform reported.