When you press Record in a meeting and choose "Record on this computer", Zoom writes the files to your disk after the call ends. On Windows and macOS the default location is a folder named Zoom inside Documents, with one subfolder per meeting named after the date and the meeting title. Each subfolder normally contains the full video as an .mp4 and a separate audio_only.m4a, and if the host enabled it, one audio file per participant as well.
Cloud recordings behave differently. They live on the Zoom web portal under Recordings, not on your computer, and Transcrify cannot read a Zoom share link because those pages require a login and often a passcode. Open the recording in the portal, use its Download button to save the MP4 or the M4A, and upload that file here. The result is the same.
If you are not sure which option the host used, look in the Documents folder first. If there is nothing there and you were the host, the recording is almost certainly in the cloud. If you were a participant, you will need the host to send you the file, because Zoom only hands recordings to the person who started them.
Zoom generates its own transcript only for cloud recordings, and cloud recording is part of the paid plans. A local recording, which is what most people on a free or basic account end up with, comes with no text at all. The same is true for calls where the host simply forgot to switch audio transcription on in the account settings before the meeting started. In both cases you have the audio and nothing else.
That is the gap this page fills. Upload the file, wait a few minutes, and you have the conversation as text, with the speakers separated and timestamps you can click through. The usual reasons people need it are minutes for a meeting they attended, quotes from a research interview held over Zoom, and lectures or webinars that were recorded for students who could not attend.
Zoom audio is a favourable case for speech recognition. Everyone speaks into their own microphone, the platform already suppresses background noise, and participants mostly take turns rather than talking over each other. The model labels each voice it can distinguish and keeps the label consistent through the call, so a one-hour discussion between four people comes back as a readable dialogue rather than a wall of text.
The language is detected automatically from the first seconds of audio, and more than fifty languages are supported, so a meeting held in German or Portuguese needs no settings. Processing takes about three to five minutes for an hour of audio because the recording is split into segments that are handled in parallel. A full-day workshop will still fit, as long as a single file stays under 3 hours.
The recording is processed on our own servers with our own models. It is not forwarded to a third-party speech API, which matters when the call was a client meeting, a hiring interview or an internal review that was never meant to leave the company.
As a guest you can read the transcript and copy it. A free account turns it into something you can work with: an editor for fixing names and product terms the model misheard, a history of everything you transcribed, and downloads as TXT, SRT, VTT or ASS. The free account comes with 5 hours of transcription across 10 files, and the transcripts you made here as a guest are moved into it when you sign up. AI chat with the recording, automatic summaries and PDF or Word export are part of the paid plans.