Dictation glossary

On-device transcription

On-device transcription means speech recognition that runs on the computer or phone in front of you, using a model loaded in local memory. Microphone audio is scored on that hardware, and the transcript is produced without requiring a remote speech API. It is the usual foundation for offline dictation. Cloud transcription is the alternative, where audio is sent to a server that returns text.

Get WhisperJot Try the browser demo
In more detail

What is On-device transcription?

On-device transcription starts with a model file on disk. At runtime the app loads it onto the CPU, GPU, or a dedicated neural accelerator and feeds it audio buffers. There is no magic privacy from the file sitting there — the gain is that recognition does not need to copy the waveform to someone else's GPU. First launch is often slow because of the download and compile step. Later launches reuse the cached model. Hardware bounds are real: machines with less memory struggle with large checkpoints, especially if you also run a local language model for cleanup.

On-device is not a synonym for 'never uses a network.' The app may check for updates, sync settings you opt into, or offer a cloud engine as a second mode. The scoped claim is about where transcription happens when the local engine is selected. It also is not a synonym for 'more accurate': a huge cloud model can beat a tiny local one on some audio, while a strong local model can beat a cheap cloud endpoint. Compare engines, not slogans.

For writers who speak

Why it matters for dictation

Dictated speech is often confidential: names, strategy, source code, health notes. On-device transcription keeps that audio on the machine for the recognition step, which is the moment most cloud tools take a copy. It is also how dictation survives a dead connection. The cost is a model download and hardware that can keep up with real time.

In this product

How WhisperJot handles it

WhisperJot is private by default: Jot Local and Jot Local Pro transcribe on-device and work offline after the first model download. Meeting transcription always uses a local model, even if you selected Jot Cloud for dictation. Jot Cloud itself is opt-in, runs only while selected, and audio is never stored on our servers.

Related terms

Keep reading in the glossary.

Cloud transcription

Cloud transcription means speech-to-text that runs on a remote server: your app uploads audio, a model in a data center produces text, and the words come back over the network.

Read Cloud transcription →

Offline dictation

Offline dictation means speech-to-text that keeps working with no internet connection because the recognition model runs on your computer.

Read Offline dictation →

Apple Neural Engine

Apple Neural Engine is a dedicated block of silicon on Apple chips that runs neural networks locally, alongside the CPU and GPU.

Read Neural Engine →

Whisper

Whisper is an open-source family of automatic speech recognition models that convert recorded speech into text.

Read Whisper →
Questions

Straight answers.

See also: System requirements, Pricing.

What is on-device transcription?

On-device transcription is speech-to-text that runs on your own hardware. A local model scores the microphone signal and emits words without calling a remote recognition API. That is what makes offline dictation possible. You still need disk for the model and enough memory to load it. Cloud transcription sends audio out instead, then returns text.

Is on-device transcription the same as offline dictation?

They overlap, but they are not identical phrases. On-device describes where the model runs. Offline describes whether a network is required at the moment you speak. A local model that still phones home is on-device but not offline. In well-designed apps, selecting the local engine gives you both: recognition on the machine, no connection needed after the download.

Does on-device transcription use the Neural Engine?

On Apple silicon it often does, through runtimes that compile models for the Neural Engine or GPU. On Windows and Linux the same checkpoint may run on the CPU or a discrete GPU instead. Acceleration changes speed and battery use; it does not change the definition. If there is no neural coprocessor, on-device still just means 'on this computer.'

One hotkey, any focused app.

Private local transcription by default, with an optional opt-in cloud engine.

Get WhisperJot — $99/year