Pandrator 0.8.7 improves subtitle and speech-block quality for long-form voiceovers, especially material transcribed with MOSS, and makes correction and translation jobs easier to control and diagnose.
What’s improved
- MOSS chunk seams are now handled without intentionally re-transcribing overlapping audio, and strong time-aligned duplicate word runs from older or imported results are removed safely during subtitle finalization.
- Subtitle boundaries now use speaker, punctuation, sentence-model, reading-speed, and silence evidence together. A 1.5-second hard-gap default avoids bridging conspicuous pauses without producing the fragmentation seen with a 1-second limit.
- Speech blocks keep exact subtitle lineage and explicit alignment groups. Long text is split at the TTS engine’s hard limit, related pieces are assembled before timing, speaker changes remain separate, and an unfinished thought can continue across ordinary cue boundaries without spanning a long silence.
- Correction and translation can include cue timings, gaps, overlaps, and speaker information for higher-quality editorial decisions. They can also run multiple independent LLM requests concurrently when speed matters more than previous-block context.
- Reasoning level, concurrency, and timing context are available from the stage editors. Correction, translation, TTS optimization, and voice-generation cards now show elapsed time, cost, token usage, and model information when available.
What’s fixed
- Gemini 3 native function calls now preserve the complete assistant message and thought signature across tool-result turns, including parallel calls. Empty provider responses also report useful finish and reasoning-token diagnostics.
- Opening full settings from a curated stage editor now keeps the unsaved model, reasoning, instruction, and other choices from that editor.
- Generation-drawer find and replace is atomic across all matching segments, handles Unicode boundaries correctly, and cannot leave a partially replaced result after a conflict.
- Generated takes update in place as work completes, so playable controls no longer depend on changing a filter to refresh.
- Sessions with subtitles but no generated voiceover now default to preserving the source audio and exporting subtitles, instead of suggesting a mixed voiceover output prematurely.
Existing sessions are migrated automatically. The new subtitle and speech-planning thresholds remain editable in the relevant settings panels.
Which file should I download?
- Windows 10 or 11 (64-bit):
PandratorManager-0.9.14-windows-x86_64.exe - Linux desktop (64-bit):
PandratorManager-0.9.14-x86_64.AppImage
The ZIP files and Python packages are intended for manual, command-line, or automated setups. Most people do not need them.
Updating an existing installation
Open Pandrator Manager and use the update button on the installation screen, or go to Maintenance → Updates and repair. Existing sessions, sources, voices, models, and service data remain in persistent storage.
This release contains Pandrator 0.8.7 and Pandrator Manager 0.9.14. The Manager binaries are unchanged from the previous release and are included again for convenient fresh installs. All download hashes are collected in the single SHA256SUMS file, and Manager update bundles remain authenticated by the signed release manifest.