Transcribe a Teams Recording Without Access to the Tenant

Where Teams stores the recording

Microsoft Teams saves meeting recordings as MP4 files, but where they end up depends on how the meeting was set up. A scheduled or ad-hoc meeting that is not tied to a channel goes to the OneDrive of the person who started the recording, into a folder named Recordings. A meeting held inside a channel goes to the SharePoint site behind that channel, in the Files tab, again under a Recordings folder. The link that appears in the meeting chat points to one of those two places.

Transcrify cannot open that link, because OneDrive and SharePoint files are protected by the tenant's permissions. Open the recording in your browser, use the Download option in the toolbar, and upload the saved MP4 here. If Download is missing, the organisation has restricted it, and the person who recorded the meeting will have to send you the file another way.

When the Teams transcript is out of reach

Teams can produce its own transcript, but only when someone turns transcription on during the meeting, and the result stays inside the organisation's tenant. External guests, which in practice means clients, contractors, candidates and partners, can often watch the recording that was shared with them but cannot open the transcript tab, because that part of the meeting lives in a workspace they are not a member of. The same happens to employees once they leave the company.

Then there are the meetings where transcription was never switched on. Recording and transcription are separate switches in Teams, and a long recording with no transcript is the normal outcome when nobody thought about it at the start. Finally, some organisations disable live transcription altogether for compliance reasons while still allowing recording. In each case the recording exists and the text does not, and that is exactly what this page is for.

What the output looks like

A Teams recording is good material for speech recognition: one microphone per participant, echo cancellation already applied, and a conversation that mostly runs in turns. The model labels each voice it can tell apart and keeps the label consistent across the file, so a steering committee with six people comes back as a readable dialogue. The language is detected automatically, more than fifty are supported, and a call that mixes English and French is handled without any setting.

Processing takes roughly three to five minutes per hour of audio, because the file is split into segments and handled in parallel. The free limit is 3 hours per file, which is enough for a full workshop. The recording is processed on our own servers with our own models and is not passed to a third-party speech API, a point worth knowing when the meeting was a board review or a negotiation.

Keeping and editing the transcript

As a guest you can read the transcript and copy it. A free account adds an editor for correcting names, project codes and acronyms the model could not know, a history of your recordings, and downloads as TXT, SRT, VTT or ASS. It comes with 5 hours across 10 files, and the transcripts you made 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.