On-device transcription means the speech model runs on your own Mac or iPhone, and the audio is transcribed there instead of being uploaded to a transcription service. Cloud transcription means the audio is uploaded to a server, processed there, and the text is sent back. The choice affects who can hear your recordings, what you pay per hour of audio, and whether it works on a plane.
This article explains the mechanics of both, does the billing math honestly, names the situations where cloud is still the better tool, and ends with a checklist for telling whether an app really transcribes on the device.
What "on-device" means technically
A speech recognition system has two parts: the model, a large file of learned weights, and the runtime, which feeds audio through those weights to produce text. In a cloud service both live in a data center. In an on-device app both live on your hardware.
Concretely, an on-device transcription app:
- Gets the model weights once, usually at first launch or in the background soon after. This is the one time it needs the network for the core job.
- Stores them on disk, typically a few hundred megabytes to a couple of gigabytes per model.
- Runs the model on the device's own chip. Apple silicon has hardware built for exactly this kind of work, which is why on-device transcription became practical on phones.
- Writes the text to the device's storage. The audio is not sent to a transcription server along the way.
The practical test is simple. Turn on airplane mode, or turn Wi-Fi off, and transcribe something. If it works, the recognition happens on the device. If the app shows a spinner or an error, it does not, whatever the marketing says.
Two phrases are often confused with on-device processing:
- "Encrypted in transit" means the upload is protected from eavesdroppers on the way to the server. The server still decrypts it and processes everything.
- "We do not store your audio" describes retention, not location. The audio was still processed on a machine you do not control, by a company that had it, however briefly.
Apple's own dictation is a useful reference point: for some languages and devices it runs on the device, and for others it sends audio to Apple's servers. Same feature, two different privacy stories, which shows why the question is worth asking of every tool.
What cloud transcription services do with your audio
A cloud service has to do the following, and each step is a place where a policy applies:
- Receive the audio. It now exists on their infrastructure, and often on a hosting provider's infrastructure beneath that, sometimes in another country.
- Process it. Larger providers may route work to subprocessors, listed, if at all, deep in a data processing agreement.
- Return the text, and often keep both for a period: to let you download again, to debug, or to "improve the service". Training on customer data is usually opt-out or off for paid tiers, but the default varies by product and changes over time.
- Delete, eventually. Retention windows in consumer apps range from immediate to indefinite.
None of this is sinister. It is the normal shape of a cloud product, and reputable services document it. But it means "is my transcript private" becomes a question about a vendor's policy, its subprocessors, its breach history and its jurisdiction. With on-device transcription, that chain of questions does not arise for the transcription itself, because no transcription vendor receives the audio. For professionals with confidentiality duties, that is the point; private medical and legal dictation goes into specifics.
The billing math
Cloud transcription is metered because every minute of audio costs the provider compute. That reaches you as a per-minute price on an API, or as a monthly plan with a cap on minutes.
Raw APIs are cheap per minute: roughly a fraction of a cent to a couple of cents, and cloud transcription API costs works through the arithmetic. Consumer apps built on top of them charge more and add a monthly allowance. To compare, work in hours of audio per month:
| Audio per month | At $0.01 per minute | At $0.02 per minute |
|---|---|---|
| 5 hours | $3 | $6 |
| 20 hours | $12 | $24 |
| 40 hours | $24 | $48 |
| 100 hours | $60 | $120 |
Someone transcribing two meetings a week is in the 5 to 10 hour band, where cloud is genuinely inexpensive. Someone transcribing every call, lecture or interview is in the 40 to 100 hour band, where the meter matters and a capped consumer plan usually forces an upgrade.
On-device transcription has no per-minute meter, because the compute is yours. Apps charge for the software instead, by subscription or one-time purchase, so the hundredth hour costs the same as the first. At low volumes a cheap metered service can cost less, and it is fair to say so.
Latency, speed and working offline
Transcribing a file in the cloud means uploading it first. A one-hour recording is tens of megabytes; on hotel Wi-Fi the upload can take longer than the transcription. Live cloud captioning adds a round trip on every phrase, which is why cloud dictation can feel a beat behind your voice.
On-device transcription starts immediately and runs as fast as the local chip allows. On recent Apple silicon, modern speech models transcribe far faster than real time, so an hour of audio finishes in minutes with no upload step. Offline speech to text on iPhone explains how that works on a phone.
And then there is having no network at all: trains, planes, hospitals, courtrooms, basements, field sites, a client's office where you do not want to join the guest Wi-Fi. On-device works in all of them. Cloud does not, and uploading later is a workaround, not a fix.
Where cloud transcription still wins
Being honest about this is more useful than pretending otherwise.
- Weak hardware. On an old laptop or a phone with little memory, a large on-device model may not run, or runs slowly enough to be annoying. A data center does not care what you own.
- Huge batches. A thousand hours of archive audio in one go is a job for many servers in parallel.
- Shared workspaces. If a team needs the same transcripts, comments and search in one web place, a cloud product provides that by design. An on-device app can sync between your own devices, but it is not a collaboration server.
- Some languages and accents. The largest cloud models are still ahead for low-resource languages and very heavy accents. For English and the major European and Asian languages the gap has narrowed to the point where most people cannot tell.
- Other platforms and big events. On Windows or Android the on-device options are thinner, and streaming captions to hundreds of viewers belongs on a server.
If you are in one of these groups, use the cloud with eyes open: read the retention policy, turn off training where offered, and do not upload anything you would not email.
How to tell if an app is really on-device
A short checklist, for any transcription app:
- Does it work in airplane mode after setup? This is the test that matters.
- Does it require an account? An app that transcribes on your device has little reason to know who you are.
- What does the App Store privacy label list? Compare it with what the app says, and check whether any diagnostics are optional.
- Does it say plainly where each feature runs? Some apps transcribe on the device but send the text to a cloud model for summaries, or use features provided by the platform. A clear app says which is which.
- Is there a per-minute meter? A minutes allowance almost always means someone else's servers are doing the work.
How to run the checklist on ThinkScribe
ThinkScribe transcribes and separates speakers on your device. Here is how to check that yourself on iPhone, iPad or Mac:
- Open ThinkScribe once while online so any one-time downloads can finish, including ThinkScribe AI if you choose it in the AI settings.
- Turn on airplane mode.
- Record a minute of speech and stop. The transcript and speaker labels are made on the device, with no connection.
- Summarize the recording. With ThinkScribe AI selected, summaries also run on your device. If you choose Apple Intelligence instead, that feature is provided by Apple.
- Note the parts provided by Apple: Apple Intelligence if you choose it, and iCloud sync, which is off by default and goes through your own iCloud account.
There is no account, and your recordings and transcripts are never sent to us. The app sends us only crash reports and usage counts, never your audio, transcripts or any other content, and you can turn that off in Settings. The complete guide to voice to text and the audio to text page cover the rest.
Frequently asked questions
Is on-device transcription as accurate as cloud transcription?
For English and the major world languages, current on-device models are close enough to cloud services that the difference is rarely noticeable in everyday recordings. Cloud models keep an edge for low-resource languages and very heavy accents, and the biggest models still do better on very noisy audio.
Does offline transcription mean my audio is private?
It means the transcription happens on your device and the audio is not uploaded to a transcription service. The test is whether transcription works with the network off. What remains is your own device's security, your backups, and any features you choose that another company provides, such as cloud sync.
How much does cloud transcription cost per hour?
Raw API pricing works out at roughly $0.40 to $1.50 per hour of audio, and consumer apps usually bundle a monthly allowance at a higher effective rate. At 20 hours a month that is around $12 to $30. An on-device app charges for the software, not per minute.
Can on-device transcription work without internet?
Yes, that is the defining property. After any one-time downloads, recognition, and in apps that also run their AI on the device, summaries too, work with no connection: in airplane mode, in a basement, or in a building where you will not join the Wi-Fi.
