A transcription app that does not upload your audio
Cloud transcription is fast and accurate because it runs on somebody else’s hardware — which means your meeting, your interview or your lecture is copied onto a server before it becomes text. Murmur keeps the entire job on your phone.
Two designs, one difference you can feel
From the outside the two look the same: you press record, a transcript appears. The difference is what happened in between. A cloud service compresses your recording, sends it across the network, queues it on a machine you cannot see, and sends text back. An on-device app loads a speech model into your phone’s memory and does the same work there. Everything else — the privacy story, the offline story, the account, the “is the service up?” question, the data-processing agreement your legal team wants to read — follows from that single choice.
- No upload step, so no upload to worry about
- There is no transfer to intercept, no storage bucket to misconfigure and no retention window to take on faith. Your recordings are never sent to our servers because there is no server in the path.
- No account to create, and none to delete
- Cloud transcription starts with a sign-up, because the server has to know whose file it is holding. Murmur has nothing to sign into: no profile, no password, no delete-my-data request to chase.
- The network stops mattering
- A hosted transcript needs bandwidth, and a bad conference-room connection becomes a spinning progress bar. On-device transcription runs at the speed of your phone, in a basement or at 30,000 feet.
- It keeps working when the service does not
- Outages, plan changes and shutdowns are the ordinary life cycle of a hosted product. A model sitting on your phone does not have a status page.
Side by side
On-device app vs cloud transcription service
| Capability | Murmur (on-device) | Typical cloud service |
|---|---|---|
| Where transcription happens | On your own device | On the provider’s servers |
| Your recording is uploaded | No | Yes — that is how it works |
| Account required | No | Usually |
| Works with no signal | Yes, after a one-time model download | No |
| Speaker labels | On-device, remembered across recordings | Server-side, if the plan includes it |
| Transcript translation | On-device, line by line | Server-side, where offered |
| Transcribe a file you already have | Yes, on-device | Yes, by uploading it |
| Export formats | Plain text, Markdown, PDF, Word | Varies by plan |
| Who else can read the transcript | Nobody unless you share it | You and the provider |
| If the service is down or shuts down | Nothing changes — it all runs locally | You lose access |
What you give up by staying local
Honesty first: an on-device model has to fit inside a phone’s memory and power budget, so a cloud service running a much larger model on a rack of accelerators can be more accurate on genuinely hard audio — a noisy room, strong accents, three people talking at once. Murmur narrows the gap where it matters instead of pretending it does not exist: tap any line to replay exactly that sentence and check the wording, reassign a line to the right speaker, or relearn a voice so the next recording comes back better. For ordinary meetings, interviews and lectures the transcript is something you can work from, and anything critical takes seconds to verify.
Why the speaker labels are the other half of this
A transcript with no attribution is still a wall of text you have to decode. Murmur splits the conversation by voice, color-codes it and remembers people between recordings, so next week’s meeting arrives with your teammates already named. That work is local too, and it matters more than it sounds: a model of somebody’s voice is biometric data, and the safest place to keep it is the device that created it.
Who should not choose the on-device one
If what you need is a shared team workspace, live collaborative editing, or an API that pipes transcripts into a data warehouse, a hosted service is the right tool and this comparison is not close. Murmur is built for the other case: one person, one device, conversations that should not be copied anywhere, and a transcript you can export to Plain text, Markdown, PDF or Word and take wherever you want — including into whichever AI assistant you already trust.
FAQ
On-device vs the cloud — common questions
Is on-device transcription as accurate as a cloud service?
Does Murmur ever send my recording anywhere?
Do I need to create an account?
Can I still get a summary out of it?
The other comparisons
Offline transcription app
What “offline” actually means when the speech model lives on the phone instead of on a server.
See offline transcription appPrivate transcription app
Why “we do not look at your recordings” and “your recordings never leave the device” are not the same claim.
See private transcription appStart with your next conversation.
Download Murmur, download a speech model once, and every recording after that is transcribed on your own device — meetings, interviews, lectures, ideas on a walk.
Free to download · No account · Works with no signal