whispem/sussurro.cpp

sussurro is an offline translation tool that runs entirely on your own computer, with no internet connection needed once everything is set up. It translates between English, Spanish, French, and Italian, and it can handle both typed text and spoken voice, in either direction. You can speak a sentence in one language and hear it read back in another, or simply type a sentence and get translated text back. The project is built from a few pieces working together. Translation is handled by a custom engine written from scratch rather than relying on an outside cloud service, using the same kind of neural network technology behind commercial translators. Turning speech into text uses a well known open source project called whisper.cpp. Turning translated text back into spoken audio uses a separate voice synthesis library that plays a natural sounding voice for each supported language. All of this runs locally, using the computer's own processing and graphics hardware rather than sending anything to a server, so conversations and translations stay private. There are several ways to use it. A command line tool lets you translate typed text directly, and separate command line tools handle turning speech into text, translating spoken audio, and turning translated text into speech. There is also a small script that demonstrates a full voice to voice loop, where you speak in one language and the program plays back the spoken translation in another. Beyond the command line, the project includes a native desktop application built with Tauri, where you pick a source and target language, then type or speak, and read or hear the translated result, with a button to quickly swap the two chosen languages. Setting it up takes a bit of work. Building the underlying engine requires a C++ build toolchain, and getting translation working means downloading and converting the specific language model files you want to use, which can be shrunk down to save disk space. Voice input needs an additional speech recognition model, and voice output needs a separate downloaded library plus one voice file per language. The desktop app additionally needs Rust and Node.js installed. The project has mainly been tested on Apple computers with Apple Silicon chips, and support for Windows, Linux, and Intel Macs is not yet confirmed.