Where meeting recordings usually come from
Meeting transcription starts with a file, and meetings leave files in many places. An in-person meeting is often recorded on a phone app or a handheld recorder, which saves an M4A, MP3 or WAV file. Google Meet saves recordings as MP4 video to the organiser's Google Drive, in a folder called Meet Recordings, on Workspace plans that include recording. Webex cloud recordings can be downloaded as MP4 from your Webex recordings list. Other video tools and conference room systems usually produce MP4 too.
You can upload any of these directly. Audio is taken out of video files on our side, so there is nothing to convert first. If the recording sits in Google Drive or Dropbox, you can also paste the link instead of downloading it, as long as the file's sharing settings let it be opened from the link. For Zoom and Microsoft Teams there are separate pages that explain where those apps put their recordings.
Recording a meeting so the transcript is usable
Most problems in meeting transcripts start in the room. A few habits make a clear difference:
- Put the phone or recorder in the middle of the table, not at one end next to the person who set it up. Distance is the main reason quiet speakers get lost.
- Keep it away from laptop fans, projectors and air vents. Steady background noise blurs consonants.
- Put the phone face up on a soft surface like a notebook, so tapping and cups set down on the table do not thump through.
- Ask people to finish their sentence before the next person starts. Two voices at once are hard for any speech recognition to separate.
- Have people say their name the first time they speak in a large meeting. It makes labelling speakers later much quicker.
- For a hybrid meeting, record from the meeting software rather than from a phone in the room. The software captures remote voices directly instead of through a speaker.
From transcript to minutes
When the file is done you get the full transcript, split into paragraphs with timestamps. A player sits under the text and highlights the line being spoken, so you can click any line to hear that moment again. Use this to check the parts that matter: decisions, numbers, dates, owners and names of products or clients. Fix any wrong word directly in the text.
On paid plans (Plus, Unlimited) or with a Minute Pack, speaker identification labels who said what. If you know how many people spoke, you can enter the number, which helps it split voices correctly. Rename Speaker 1 to "Priya" once and every line she spoke updates. Speaker labels are not part of the Free plan or the no-sign-up trial, but you still get the full timestamped transcript there.
If you want a summary or a list of action items, use the AI prompt export. It copies the transcript together with a ready-made prompt that you paste into ChatGPT or Claude. RocketScribe itself does not write summaries, so the assistant you choose does that step, and you can check its output against the transcript.
Which export fits a meeting
Pick the format by what happens next. Speaker names and timestamps can each be turned on or off when you export.
- DOCX (Word) for minutes you will edit, tidy and send around. Turn timestamps off for a cleaner document.
- PDF for a fixed record to attach to an email or file with the meeting papers.
- TXT or the Copy button to paste into a wiki, ticket or chat message.
- A read-only share link for colleagues who missed the meeting and want to read or search it themselves.
- CSV if you track decisions in a spreadsheet and want each line with its timestamp in a row.
- SRT or VTT if the meeting recording will be published as a video with captions.
Honest limits for meetings
Meetings are one of the harder kinds of audio. People interrupt each other, laugh, eat and speak from across the room. Clear speech close to the microphone transcribes very well. Crosstalk, side conversations and voices far from the recorder come out less reliably, and when people talk over each other speaker labels can switch to the wrong person for a line or two. Plan a short review pass for anything that will be shared as an official record. Uploads are never used to train AI models, and you can delete a recording and its transcript whenever you want.