Transcribe speech to text instantly on macOS using mouse buttons or shortcuts
Press a mouse side button or a keyboard shortcut, and your spoken words are converted to text in real time. The app uses a local speech recognition model, allowing you to quickly turn personal voice notes into text without an internet connection. It is optimized for Macs with Apple Silicon.
Flexibility in recording methods
Users can start and stop recording by pressing a global shortcut or a mouse button (middle or thumb button). A hold mode is supported, which records only while the button is pressed, allowing you to capture specific segments accurately. Microphone selection is flexible, with the ability to switch between the built-in microphone, external microphones, Bluetooth, and the iPhone's Apple Continuity microphone directly from the menu bar.
Two engines and language support
The app provides two speech recognition engines, Whisper and Parakeet, and allows you to download models directly within the application.
- Model management You can use Whisper model files (.bin) by downloading them from Hugging Face and placing them in the app's model directory. A default model is copied automatically on first launch, but additional models must be managed manually.
- File processing Recorded audio files can be imported via drag and drop and converted in bulk using a queue processing method.
Installation and system requirements
Install via Homebrew using the brew install opensuperwhisper command, or download the binary from the GitHub releases page. The app runs only on Apple Silicon (ARM64) based macOS and does not yet support Intel-based Macs. Building locally requires CMake, libomp, Rust, and Ruby, and you can configure the build environment by referring to the CI workflow.
Current limitations and future plans
Features such as streaming conversion, Intel macOS compatibility, agent mode, and background app execution are currently in development or planning stages. When reporting bugs, it is recommended to create an issue including system information and logs. The project is distributed under the MIT license, allowing free use and modification.