Apache 2.0 · Built in public

Whisper Desktop

A universal voice keyboard for clean speech-to-text in any desktop application—without a subscription or platform lock-in.

View and star on GitHub →

Dictate anywhere

Use one global hotkey to capture speech and insert the result into the field where you started.

Edit by voice

Select text, speak an instruction, and replace it with a rewritten version.

Custom vocabulary

Bias transcription toward names, acronyms, and terminology that matter to you.

Hands-free option

An optional local wake phrase starts and stops dictation while ambient listening remains on the device.

Failure recovery

Failed recordings are preserved for retry instead of disappearing when an API call fails.

Operational visibility

Settings, CLI controls, persistent logs, and latency summaries make the pipeline inspectable.

Why it exists

Voice input should be a capability, not a subscription.

Whisper Desktop began as a weekend tool for speaking into any application on the computer. It grew into a more resilient dictation pipeline with voice commands, configurable models, clipboard restoration, long-recording support, and recovery when transcription fails.

The repository includes the application, CLI, tests, architecture notes, contribution guidance, and build instructions. Issues and pull requests are welcome.