Built a local dictation app in Rust/Tauri that runs Cohere Transcribe (#1 open A

On April 13th, a developer posted to LocalLLaMA describing a dictation application built in Rust using the Tauri framework, running Cohere Transcribe entirely on-device — no cloud handoff, no subscription fee, no audio data leaving the machine. The model choice is deliberate rather than decorative: Cohere Transcribe currently sits at number one on the open automatic speech recognition leaderboard, posting a 5.42% word error rate. That figure matters because it was earned through measurement and ranking against alternatives, not through marketing copy. The project is open source. That last detail is the one that turns a personal tool into a community artifact. One developer built something for themselves; the open-source release is an implicit invitation for others to build on top of it, tear it apart, or simply use it without asking anyone's permission.

If confirmed, here is what this means. The practical ceiling on local speech recognition just moved. For years, the honest answer to "can I run serious dictation locally?" was "sort of, if you don't mind the errors." A 5.42% word error rate from the top-ranked open model, running inside a Rust application with Tauri's comparatively lean overhead, suggests the answer is now closer to a clean yes — at least for English, at least on capable consumer hardware. The people most immediately affected are the ones with the strongest reasons to keep audio off someone else's servers: journalists protecting sources, clinicians constrained by patient privacy rules, lawyers, researchers, anyone conducting sensitive interviews. Beyond privacy, there is a cost argument. Subscription dictation services are not expensive until they are, and an open-source local alternative with competitive accuracy changes the negotiating position of anyone currently paying. Second-order, this puts pressure on cloud-first transcription products to justify their data collection in terms of quality improvements that a locally-running leaderboard model increasingly does not require.

Watch for other developers forking or starring the repository in meaningful numbers — that would signal community validation rather than a single polished demo. Also watch for independent accuracy tests on hardware representative of what non-developers actually own.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →