Frequently Asked Questions

OpenASR is a local-first speech-to-text project with macOS and Windows desktop apps, a public Rust CLI, a local HTTP API, a signed model catalog, and an Apache-2.0 open core. Transcription runs on hardware you control.

By default, no. Transcription runs locally. Network exceptions include explicit model downloads, desktop update-manifest checks, an optional Hugging Face token sent only to huggingface.co, optional online text processing that sends recognized text—not audio—to the selected provider, and optional Remote Compute—which sends audio only to a user-controlled device running OpenASR Server that you explicitly pair.

Not for transcription itself. An internet connection is needed to download models, check for desktop updates, use optional online text refinement or translation, or use optional Remote Compute. The CLI --offline option forbids network access and fails if a model is missing.

No. No account or sign-in is required to transcribe. There is no telemetry or usage analytics.

The desktop apps run on macOS and Windows. The CLI and server also run on Linux through prebuilt releases or source builds. A native Linux desktop app is not published.

Start with your language and task, then consider your hardware and memory, then required features like timestamps, punctuation, translation, or diarization, then license. The catalog marks a recommended quantization per model. Visit the live Models page for current availability.

The core engine — Rust CLI, server, and supporting libraries — is Apache-2.0 open source. The macOS and Windows desktop apps are separate closed-source products built on the open core.

Models come from a signed and verified catalog. Normal transcription does not silently download a missing model; the user sees and approves every download. An optional user-supplied Hugging Face token is sent only to huggingface.co for eligible downloads.

Voice ID lets you enroll a short voice sample so local file transcripts can show a display name instead of only Speaker 1 / Speaker 2. It works with every local ASR model on Desktop and is a labeling aid, not authentication.

Yes. Desktop supports speaker separation and optional Voice ID with every local file ASR model: it keeps native tracks when available and uses a shared external pipeline otherwise. The CLI and server also expose anonymous --diarize and operator-only Voice ID APIs.

Transcription itself runs locally once models are installed. Use CLI --offline to forbid network access. Optional online text refinement or translation and optional Remote Compute require a connection, but neither is needed for local transcription.

OpenASR runs local model packs from a signed catalog that includes Whisper-class and other families, with a desktop app on macOS and Windows plus a local OpenAI-compatible API. There is no central cloud transcription backend.

Yes. Pick a Chinese-capable or multilingual pack from the signed catalog and transcribe on your machine. No cloud account is required for local transcription.