OVR Overlay Translator · beta, public · first seen
This update focuses on smoother and more reliable spoken translations, framed screen translation on HDR displays, and a few everyday rough edges.
New
Long passages are now read in segments: they are split at punctuation and played in order, so the first words come out without waiting for the whole passage to be synthesised
The settings page reports skipped sentences when synthesis can't keep up ("Speaking backlog: %d sentence(s) skipped")
Improvements
Continuous speech sounds smoother: consecutive sentences reuse the same playback stream, so short sentences no longer pause between them
The speech model now stays loaded for 3 minutes of inactivity instead of 30 seconds, so occasional use no longer pays a reload every time
Easier diagnosis: when nothing is heard in a game, the log now records the playback device actually used and whether it fell back to System Default
Fixes
Framed screen translation in PC mode came out overexposed with washed-out shadows on HDR displays, lowering recognition accuracy
Speech stopped working entirely after a period of inactivity, with no message in the UI, and only restarting the app brought it back
Split long passages could make the model keep writing on its own, because the last segment had no closing punctuation
A selected playback device with a long name (truncated by the system) was shown as System Default in the settings page
Sentences that waited too long in the queue are no longer read out, so you don't hear content that was finished long ago
| Depot | New manifest | Previous manifest |
|---|---|---|
| 4304621 | 7140530733644404643 | — |