SelenaCore Update: High-Speed Offline Voice Recognition with Vosk Migration

I am proud to announce a major architectural milestone for SelenaCore. We have completed a full migration of our Speech-to-Text (STT) engine from Whisper to Vosk (40fc5ee). This transition enables lightning-fast, offline streaming audio processing, meaning the system now understands your commands as you speak them, without waiting for the end of a sentence.

To support this shift, we’ve introduced a comprehensive Vosk Management Suite. This includes a new model catalog and API for downloading and activating different languages (a2e7134), alongside a dedicated UI featuring a live audio monitor and WebSocket proxy (8cbb219). These tools allow users to visualize exactly how the system 'hears' and interprets sound in real-time.

On the hardware integration side, we have implemented an ALSA dmix mixer (a3b707a). This technical improvement allows SelenaCore to play media and system voice responses simultaneously without audio device conflicts. Furthermore, our new database-backed prompt store (9128119) makes all eight core system prompts fully editable, giving you total control over how the AI behaves and responds.

Finally, we’ve added an audio preprocessor and 'fuzzy' wake-word matching (cfae427), ensuring the system remains responsive even if your voice command isn't perfectly clear. For more details, visit our documentation.