Record speech as clearly as possible
Place the microphone near the speaker and away from fans, music and rubbing clothing. Ask people to avoid talking over one another. Before a long session, make a short test recording and listen back with headphones. These steps help capture understandable speech; they do not guarantee a perfect transcript.
Check length and file size separately
The bot accepts up to 250 MiB and 3 hours per recording, with a minimum duration of 30 seconds. A short recording can still be too large. If it exceeds a limit, prepare smaller audio files locally. Do not just rename the extension to reduce size. Keep the original before making changes.
If you split the recording, keep a simple index
Split at a pause when possible, not in the middle of a word. Name parts in order, for example lecture-01 and lecture-02. Write down where each part starts in the original. Treat timestamps in each returned document as belonging to that part; use your index when returning to the full recording.
Check the cost of separate files
Billing is rounded up separately for each recording: 40 Stars per started 20 minutes. At the current rate, one 60-minute file costs 120 Stars; three files of 20:01 cost 240 Stars in total. Avoid unnecessary splitting and check each quoted price before paying. The sample is not deducted.
Use the sample as a check, not a promise
If your allowance remains, check up to 60 seconds before ordering. The limit is 1 per day and 3 in total. Speech later in the recording may be quieter or noisier than the sample. Select the correct spoken language and plan time to review important sections.
Review the document alongside the original
Use the approximate timestamps to navigate, then listen around the passage. Check names, numbers and quotations before reusing them. The transcript is editable: correct recognition errors in your Word editor and keep a copy of the original audio until the work is finished. Questions about processing or data retention go to /support in the bot.