Offline speech to text on your desktop
In local mode OiPer does everything on your computer: it records your voice, transcribes it with a Whisper model you downloaded, and types the text into the app you are using. Nothing is sent anywhere, so it works on a plane, behind a firewall or with Wi-Fi switched off. Here is exactly what works offline, what does not, and how to check.
Updated
What works without internet
Everything you need for everyday dictation runs on your machine.
Transcription
A local Whisper model runs on your CPU or GPU. Once it is downloaded, no connection is needed and no audio leaves the device.
Read about ModelsTyping into any app
Hotkeys, profiles, auto submit and inserting text at your cursor all happen locally.
Read about ProfilesSnippets and history
Snippets expand on your machine, and every dictation is saved to a local history with audio you can play back.
Read about HistoryFormatting, with a model on your machine
Formatting in local mode runs through a language model you register. Point it at an OpenAI-compatible server on your own computer and the transcript stays local too.
Read about FormattingWhat needs a connection
These are optional, and each one only connects when you use it.
Downloading a model
Each Whisper model is downloaded once from Hugging Face. Download it before you go offline.
Read about ModelsYour own cloud provider
In cloud mode the recording, and the transcript when formatting is on, go to the provider you configured with your API key. A cloud formatting model in local mode receives the transcript.
Read about Security & Data HandlingOiPer hosted models
On paid plans, oiper mode sends the recording to OiPer’s service along with the focused app’s name, window title and text read from that window.
Read about Security & Data HandlingUpdates
The app contacts GitHub when you check for or download an update from Settings.
Read about SettingsCheck it yourself
You do not have to take privacy on trust.
Turn off Wi-Fi and dictate
Set a profile to local mode, disconnect from the internet and dictate. If text appears, nothing needed a server.
Read about ProfilesOne folder per computer
Settings, history, audio, models and logs live in %APPDATA%\com.oiper.desktop on Windows, ~/Library/Application Support/com.oiper.desktop on macOS and ~/.config/com.oiper.desktop on Linux.
Read about InstallationNothing in the background
A transcription request only happens in response to a dictation you started with your hotkey.
Read about Security & Data HandlingDelete it whenever you like
Clear history from the History page, remove models from the Models page, or delete the data folder to reset the app completely.
Read about Security & Data HandlingQuestions, answered
What people ask before they start dictating with OiPer.
Is there speech to text that works without internet?
Yes. OiPer transcribes speech on your own computer with local Whisper models, on Windows, macOS and Linux. After a model is downloaded, dictation keeps working with no internet connection.
Does OiPer send my audio anywhere?
Not in local mode. Audio is only sent if you choose a cloud provider with your own API key or OiPer’s hosted models for a profile.
Do I need an account for offline dictation?
No. Local transcription is free, unlimited and needs no account. An account is only needed for OiPer’s paid hosted models.
Can I clean up the text offline?
Yes, if the formatting model runs on your own machine. Formatting in local mode uses a language model you register, and any OpenAI-compatible endpoint works. A cloud model would receive the transcript.
Where are my recordings stored?
Recordings are kept in the audio folder of OiPer’s data directory on your computer, and transcripts in a local database. Neither is uploaded.
Is Windows voice typing offline?
No. Microsoft describes Windows voice typing (Win + H) as online speech recognition that needs an internet connection. OiPer’s local mode transcribes on your PC instead.
Does offline dictation need a powerful computer?
No. Smaller Whisper models like Base run well on everyday hardware, and OiPer uses your GPU automatically when one is available. Larger models trade speed for accuracy.