Transcription Quality & Editing — Quill Documentation | Quill
A transcript is the full text of everything spoken in a meeting, time-stamped and labeled by speaker. Transcription runs locally on your device; audio is never uploaded.
To open a transcript: click a meeting in your meeting list, then open the Transcript tab beside the notes.
What makes Quill's transcripts accurate
Separate streams for mic and system audio
Quill captures your microphone and the rest of the meeting's audio as two independent streams, then transcribes each separately. This avoids the echo and cross-talk artifacts that come from transcribing a mixed audio track.
Multilingual and mixed-language meetings
Quill supports 29 languages and is tuned to handle meetings where speakers switch between them, so a meeting that swings between English and Spanish stays accurate without manual toggling.
Pick the languages you actually speak in meetings. Go to Settings → Profile → Input Language and select them. Most people pick 2-4.
Supported languages: Arabic, Cantonese, Croatian, Dutch, English, Estonian, Finnish, French, German, Greek, Hebrew, Hindi, Icelandic, Italian, Japanese, Korean, Mandarin Chinese, Norwegian, Persian, Polish, Portuguese, Punjabi, Romanian, Russian, Spanish, Swahili, Swedish, Turkish, Ukrainian.
Transcription model
Quill auto-selects from a set of transcription models, primarily different sizes of Whisper, to find the one that best fits your device, balancing accuracy against your CPU and GPU. If the chosen model runs too slowly or hits a GPU error, Quill can fall back to a different model mid-meeting.
To override the default, go to Settings → AI. You can point Quill at a different transcription model (on-device for full local control, or off-device for a faster or larger cloud model).
Editing a transcript
Two ways to correct text:
- Type directly in any block. Click into the transcript and edit it like a document. The change is saved to that meeting only.
- Highlight a phrase and replace it. Select the misheard text, choose Replace, and enter the correct version. Quill fixes it everywhere in the meeting (transcript, notes, title, participants) and remembers the substitution so future meetings on this device auto-apply the same correction. Useful for product names, jargon, and acronyms that are consistently misheard.
For speaker corrections, see Speaker Recognition & Labeling.
Search
Cmd+Fopens find-in-meeting. Use it to jump to a word or phrase inside the open transcript or notes.Cmd+Kopens global search across all your meetings and contacts.
Export the audio
- Open the meeting.
- Click the Export button on the audio timeline at the bottom.
- Choose a save location.
Delete a transcript
Delete the meeting from your meeting list. The transcript, notes, and audio go with it.
Privacy
Your meetings are private by default. Each one stays on your device unless you explicitly choose to share it.
- Audio files stay on your disk.
- Transcripts stay in Quill's local SQLite database.
- Edits, corrections, and the substitutions Quill learns from your edits all stay local.
- Only AI-generated notes leave your device, and only to the LLM provider you've chosen (or none, if you've configured a local model).
Common questions
How long does transcription take? A few minutes for a one-hour meeting. Faster than real-time on modern hardware.
How can I improve transcription quality? Quill transcription is accurate but not perfect. To improve quality:
- Use headphones and a microphone
- Meet in a quiet space
- If you are on Windows, ensure transcription is happening on your GPU in settings and that you have the latest drivers for your GPU
The transcript is missing other speakers. Most often this means Quill captured only your microphone, not the other side of the call. Check system audio permission. → Getting Started