✨ What's new
More resilient on-device model downloads (Local Inference) — On slow or restricted networks, downloading local models no longer hangs forever: stalled connections now time out and retry automatically. And if the model CDN is unreachable entirely, you can now import the model files manually instead. Local model storage is also more robust and recovers cleanly across app updates.
New meeting platform — Yandex Telemost — The browser extension now supports real-time translation on Yandex Telemost, including international Telemost hosts.
On-device voice control in the TTS card — When using local text-to-speech, you can now pick the voice directly inside the selected TTS provider card, with per-engine options.
🛠 Fixes & polish
- Sokuji no longer auto-selects a virtual audio device as your default microphone.
- The live subtitle window now correctly stays always-on-top.
- On Linux, virtual audio device failures now surface in the UI instead of failing silently.
- More reliable noise-suppression startup.
- Fixed a Kizuna AI provider crash (WebRTC advertisement / provider lookup).
Full Changelog: v0.31.1...v0.32.0