On device and cloud
Where each stage of a ScribeKey dictation runs, on your phone or with your cloud provider, and what each choice costs you.
Every dictation in ScribeKey passes through up to three stages: Speech turns audio into words, Cleanup fixes the words, and a Skill rewrites them when you choose one. Each stage decides on its own whether it runs on your phone or with a cloud provider you set up. This page explains the choice for each stage and what it changes.
Speech
Speech runs on your phone by default, with a speech model you download once. After the download it works with no network and costs nothing per dictation. A larger model usually hears better and needs more of the phone’s memory. Models lists each model with its languages, size and the memory it needs.
Cloud speech sends the recording to a provider you choose, such as OpenAI or Groq, with your own API key. You pay the provider for each request. Use it when your phone cannot run a model that hears you well enough, or when you need a language no downloaded model covers. Set it up with
Cleanup
Cleanup always runs on your phone. Basic cleanup uses fixed rules and needs no download. Refine uses a model on your phone, on phones that can run it. Neither sends anything. How cleanup changes your words shows both with examples.
Skills
A Skill rewrites text or makes a Document, such as a summary, from a session. Two Skills can run on your phone: Summary and Action items, with the notes model from
Every other Skill is a Cloud skill. It sends the text to your own provider, and only when you choose the Skill for that dictation or session. ScribeKey never runs a Cloud skill by itself. Turn Cloud skills on or off with
Finding speakers in a meeting
After a meeting, ScribeKey can work out who spoke when. It does this on your phone with the speaker model, or in the cloud with your provider if you choose that under
What happens when a stage cannot run
A stage that is not ready does not quietly move to the other place:
- Refine that is not ready runs Basic cleanup, and keeps Refine as your choice for when its model is ready.
- A Cloud skill with no working provider does not run, and your text stays as it was.
- A cloud stage never switches to the phone, and a phone stage never switches to the cloud, without your choice.
The one exception is
Choosing
Start on the phone. Download the speech model that setup suggests, and keep Basic cleanup. Add a cloud provider when you want a Cloud skill such as Email or Polish, or when on-device speech misses words you need. Use your own cloud provider shows how.
Related pages
What leaves your phone
Which of your recordings, words and settings stay on your phone, which leave only when you choose, and which requests ScribeKey makes on its own.
How cleanup changes your words
What Basic cleanup and Refine do to a dictation before it is typed, with real examples of each, and what neither of them will change.