Dictation glossary

Whisper

Whisper is an open-source family of automatic speech recognition models that convert recorded speech into text. The models are trained on a large mix of audio and can handle many languages, accents, and noisy conditions. Independent implementations run on a laptop, a server, or accelerators. Size variants let you trade accuracy against speed and memory, from tiny models to large ones.

In more detail

What is Whisper?

Whisper popularized the idea that a single multilingual model could be downloaded and run without calling a hosted API. Under the hood it maps audio spectrograms to text tokens, often in 30-second chunks for long files. That chunking is why long-form transcription can feel batch-like even when a product shows progress. Fine-tuned or distilled descendants exist, but the public family is still the reference people mean when they say they run Whisper locally. It is a model, not a dictation app: you still need capture, a hotkey, and a way to insert text.

Because the weights are open, vendors wrap them in different runtimes — some optimized for Apple silicon, some for CPUs, some for servers. Those wrappers change latency and battery use more than they change the core idea. Whisper is often slower than specialized streaming models on short bursts, and stronger on messy, multilingual, or long audio. Word error rate varies by domain: read speech scores better than overlapping conversation. If a product says it uses Whisper, ask whether it runs on-device, which size, and whether a cloud fallback is the actual default.

For writers who speak

Why it matters for dictation

Dictation apps needed a recognizer that individuals could run without standing up a speech team. Whisper made that practical, which is why so many local tools mention it. For a writer, the model name matters less than whether the app uses it on-device by default, how large the download is, and whether short utterances feel snappy enough to replace typing.

In this product

How WhisperJot handles it

WhisperJot's Jot Local Pro engine is the higher-accuracy on-device option and is built on the open-source Whisper family; the Jot Local Pro models range from about 80 MB to 1.5 GB. Jot Local stays the fast default for short utterances. Both local engines run on your device by default. Jot Cloud is opt-in if you want a network engine instead.

Questions

Straight answers.

What is Whisper in speech-to-text?

Whisper is an open-source speech recognition model family that turns audio into text. People run it on laptops and servers because the weights can be downloaded and used without a hosted API. There are multiple sizes so you can trade accuracy for speed. It is a recognizer, not a complete dictation product: capture, cleanup, and typing into apps are separate layers.

Is Whisper the same as a dictation app?

No. Whisper is the speech model. A dictation app adds a microphone pipeline, a trigger, text insertion, and usually cleanup. Many apps embed a Whisper-family model as one engine among others. Saying an app uses Whisper tells you about the recognizer, not whether audio stays on-device, whether it works offline, or how text reaches the focused window.

Does Whisper work offline?

The model can, if a local runtime has already downloaded the weights and has enough memory to load them. Offline use is a property of that runtime, not of the architecture in the abstract. A hosted demo that calls Whisper in a data center is not offline. Check whether the app stores a model file on disk and whether recognition still runs with the network off.

One hotkey, any focused app.

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