How to Transcribe a Meeting Recording to Text, SRT, or JSON
To transcribe a meeting recording, drop the file — MP3, MP4, MOV, WAV, M4A, or WebM — into a batch transcription tool, let the speech engine process it with speaker labels, and export the result as TXT, JSON, SRT, or CSV depending on what you'll do with it next. With Optimus Transcriber this runs free in your browser on Deepgram's $200 signup credit, and recordings under 20 minutes come back near-instantly.
A recorded meeting that never becomes text is a memory, not an asset. You can't search it, can't hand it to an agent, can't turn it into follow-ups or a spec or a decision log. The whole job here is converting a dead file into working material — and the conversion takes minutes.
Step 1: Get the recording into a supported format
You almost certainly already have one. Zoom, Meet, and phone voice memos all produce files the transcriber accepts directly: MP3, MP4, MOV, WAV, M4A, WebM. Video is fine as-is — you do not need to strip the audio track first. If your recording is somewhere exotic, exporting to MP3 or M4A from whatever produced it is the universal escape hatch.
Step 2: Set up the transcriber (once, about two minutes)
- Open optimustranscriber.com in your browser. There's no account to create.
- Get a free Deepgram API key — new accounts include $200 in credit, roughly 20,000 minutes of transcription.
- Paste the key into the key field. It's stored locally in your browser and sent nowhere except to Deepgram itself.
Worth pausing on the architecture, because meeting recordings are sensitive by definition — deal terms, personnel, strategy. The transcriber is 100% client-side: no Optimus backend, no server-side storage, and every request carries Deepgram's model-improvement opt-out flag, so your audio isn't retained or used for training. The recording goes from your machine to the speech engine and back. That's it.
Step 3: Drop the file and let it process
Drag the recording into the file queue. Processing time scales with length:
| Recording length | What to expect |
|---|---|
| 0–20 minutes | Near-instant |
| 20–45 minutes | May take 2–3 minutes |
| 45+ minutes | Split the file — two halves process faster and more reliably |
Diarization (who said what), smart formatting, punctuation, and utterances are on by default, so what comes back is a readable, speaker-labeled transcript — not a wall of lowercase soup. You can queue multiple files and bulk-export the lot.
Step 4: Pick the right export format for the job
This is the step people get wrong, because the format is really a decision about what happens next:
TXT — for humans
Plain text with speaker labels. Paste it into a doc, an email, or an AI chat ("summarize the decisions and owners from this meeting"). If you just want to read or prompt with the meeting, TXT is the answer.
SRT — for video
SubRip subtitle format: numbered blocks with timestamps. If the recording is going anywhere as video — a training library, a client portal, YouTube — SRT gives you captions that any player and most platforms accept natively.
JSON — for agents and pipelines
The structured option: text plus timestamps, speakers, and utterance boundaries in machine-readable form. If the transcript is input for something automated — an agent that drafts follow-ups, a script that logs decisions to your CRM, anything in a FAST-style agent stack — JSON is the format that carries all the signal. This is the difference between "a transcript" and "structured output the next agent can act on instantly."
CSV — for spreadsheets
Rows you can filter and sort. Useful when you want to analyze meetings as data — talk-time by speaker, scanning utterances across a batch of calls.
Step 5: Do something with it (the part that pays)
The transcript is the raw material, not the finish line. The highest-leverage next moves, in rough order of value per minute spent:
- Paste the TXT into an AI model and ask for decisions, owners, and deadlines. Thirty seconds, and the meeting has minutes nobody had to write. (If you're new to prompting by voice and transcript, start with how to dictate prompts to AI agents.)
- Feed the JSON to an agent that drafts the follow-up email in your voice, before you've left the parking lot.
- Archive the TXT with the project, so "what did we agree in March?" is a search, not an argument.
What about live meetings, not recordings?
The same tool has a live-mic mode: streaming transcription over WebSocket with no length limit, words appearing as they're spoken. It transcribes what your microphone hears — it doesn't join your calls as a bot, by design. If you want a meeting-bot-and-archive product, that's a different tool doing a different job; we compared the options honestly in the free transcription tools roundup.
FAQ
What file formats can I transcribe?
Optimus Transcriber's file-drop mode accepts MP3, MP4, MOV, WAV, M4A, and WebM — audio or video. You don't need to extract the audio track from a video first; drop the file in as-is.
Which export format should I choose — TXT, JSON, SRT, or CSV?
TXT for reading and pasting into docs or AI tools; SRT for captions and subtitles on video; JSON for feeding agents, scripts, and pipelines (it carries timestamps, speakers, and structure); CSV for spreadsheets and analysis. You can export the same transcript in multiple formats.
Can it tell who said what in the meeting?
Yes. Speaker diarization is enabled by default, along with smart formatting, punctuation, and utterances — so the transcript comes back labeled by speaker rather than as one undifferentiated wall of text.
How long does transcription take, and is there a length limit?
Recordings up to about 20 minutes come back near-instantly. Files in the 20–45 minute range may take 2–3 minutes to process. Past 45 minutes, splitting the recording is recommended. Live-mic streaming has no length limit.