neo773/smart-whisper

Smart Whisper lets JavaScript and Node.js developers add AI-powered speech-to-text transcription to their applications. Instead of relying on a cloud API for converting audio to text, you can run the transcription engine directly on your own machine or server, which gives you more control over the process and helps avoid ongoing usage fees. Under the hood, it acts as a bridge between your JavaScript code and a widely used open-source transcription engine called whisper.cpp. Normally, connecting JavaScript to a low-level engine requires some heavy lifting, but this package handles that connection for you. It also includes a model manager that automatically downloads the necessary AI files and keeps them updated. To save memory and computing power, it can load a transcription model just once and use it for multiple audio files at the same time, or automatically free up resources when a model is no longer needed. A developer building a podcast platform, for example, could use this to automatically generate written show notes from audio recordings. A startup creating a meeting recorder could use it to transcribe conversations for searchable archives. It is particularly useful for any project that needs to process audio locally, without sending private or sensitive recordings to an outside service. Out of the box, the package is optimized for Mac computers, taking advantage of both the main processor and the graphics chip for faster performance. On Linux and Windows, standard installation works immediately, but if you want to use a graphics card to speed up transcription, you have to supply your own compiled version of the whisper engine. This setup gives developers flexibility but means that achieving maximum hardware acceleration on non-Mac systems requires a bit of extra technical work.