## v1.1.031 — 2026-05-27
- improvement: **HQ Voice (voice-cloning) engine now covers 600+ languages**, up from the previous 10. Memory and VRAM footprints are smaller, audio output is cleaner, and voice-clone reproduction is noticeably more stable across long sentences and multilingual replies.
- improvement: **Speech recognition accuracy for European languages is substantially higher.** Russian, German, French, Spanish, Italian, Portuguese and the rest of the European set get large gains in transcription quality. Chinese / Japanese / Korean continue to use the Fast Recognition model and are unchanged.
- improvement: **HQ Voice no longer needs a manual service-start step or service-status indicator.** The engine starts on demand the first time you trigger speech and releases its GPU memory cleanly when idle — the previous start / stop button and service-state panel have been removed, so synthesis is always one click away.