Offline speech-to-text
Offline speech-to-text that keeps your voice on your machine
OpenBroca can run speech recognition entirely on-device, so your audio is transcribed locally and never sent to the cloud. It's free, open source, and works across macOS, Windows, and Linux — with cloud models always available when you want them.
What is offline speech-to-text?
Offline (or on-device) speech-to-text converts your voice into text using a model that runs locally on your own computer, rather than streaming audio to a remote server for processing. The practical difference is large: with cloud dictation your spoken words travel to a third party; with offline dictation they stay on the machine in front of you. For anyone handling confidential work — legal, medical, journalistic, or simply private — that boundary matters.
Why use offline dictation?
Audio never leaves your device
On-device recognition means your voice isn't streamed to a server — nothing to intercept, log, or retain.
Works without internet
Dictate on a plane, in a secure facility, or on a flaky connection. Local models don't need a network.
No per-minute cloud cost
Once a model is downloaded, transcription runs on hardware you already own — no metered API usage.
Cloud is still one click away
Prefer maximum accuracy or speed for a task? Switch to a cloud provider per workflow — you decide.
How OpenBroca runs speech-to-text locally
OpenBroca uses Sherpa-ONNX to run modern speech-recognition models directly on your device — download a model once and dictate with no network round-trip. Recognition is exposed through the same open interface as cloud engines like Deepgram, so switching between local and cloud is a setting, not a migration. Any API keys you do use for cloud providers are stored in OS-backed secure storage, and because the whole app is open source, you can confirm in the code that local means local. Recognized text is dropped straight into whatever app is focused, and you can route it through a language-model step to clean up or reformat it afterward.
Offline vs cloud: when to use each
Local models give you privacy and offline availability and are more than good enough for everyday dictation. Cloud models can offer an edge in raw accuracy, latency, or breadth of language support, at the cost of sending audio off-device and (usually) paying per use. OpenBroca's point is that you shouldn't have to pick once and live with it — choose local when privacy wins, cloud when capability wins, per task. Learn more about OpenBroca's open-source approach to dictation.