LiveCap

• Uninstalling removes them too — no more hunting for 37 GB by hand. This is the headline of this release • The only things it removes are the models and caches — your transcripts, your settings and your logs are never touched, and neither is anything you put in data yourself • WhisperS2T and Qwen3-ASR have not caught up yet — those two still keep their models in the old spot (inside your user profile), and uninstalling leaves them there. Sorry about that — a future update will bring them in line! ReazonSpeech and Parakeet are confirmed to run from the new location If you already have models downloaded, nothing moves. The new location only applies to fresh installs, so uninstalling leaves the old folder alone. Automatic migration from the old location is coming — please bear with us a little longer. If you want to switch right now, delete the old folder (AppData, under PineLab/LiveCap/Cache) before launching. 💬 When live translation stops, it tells you • Live translation failures now show up immediately — LiveCap used to fail and then just display the original text, which looked exactly like "I asked for a translation and got the same language back." Now you get a message such as "Cannot reach the translation service. Showing the original text only." • The message depends on the reason — unreachable service, unsupported language pair, engine failed to initialize, model failed to load. What you should do next differs, so we do not lump them all into "translation failed" • It goes away when things recover — the notice clears and you get a single "Translation recovered" message. And it will not nag you repeatedly while the failure persists • Startup failures no longer wait for your first utterance — if the translation engine fails to initialize, you used to learn about it only when the first line came through 🔧 Changes & Fixes • Fixed engines failing to load when your username contains non-ASCII characters — with a non-ASCII Windows username, ReazonSpeech, Parakeet and Canary simply could not load their models. Our logs show this hit 421 times. Renaming your user account or manually pointing LiveCap at a different model folder is no longer necessary. (Verified for those three engines; the remaining engines are still being checked) • Fixed translation returning "Japanese to Japanese" — since late August, some of you reported translated lines coming back as the original text. The translation service had started rejecting LiveCap's requests while pretending to succeed and returning an error in the body. We fixed how we talk to it, and it got a little faster on the way (403ms to 191ms, measured) • Fixed a crash when closing the app after using translation — if a translation was still running in the background at shutdown, LiveCap could occasionally terminate abnormally. Those workers are now always reclaimed on exit. As a bonus, results from the previous configuration can no longer leak in right after you change a translation setting • Rebuilt how ReazonSpeech identifies its cache — it could mix up caches from different settings, or keep using a stale cache after the model files were swapped • Fixed downloads deleting another download's temporary files — a quietly nasty one that could bite when fetching several models at once • File transcription now makes up to three attempts after a temporary network error — the first attempt plus two retries. Fewer lines dropped to a momentary hiccup • The startup log now records version, build and storage location — "which build are you running?" was the single most awkward question in support. If you report a problem, sending those first few lines along will get us to the cause much faster If something breaks Something left over after uninstalling, an engine that will not load, translation acting strange, or you just want to chat — anything goes on the official Discord. https://discord.gg/hdSV4hJR8Y Happy captioning!