• Translation survives brief network drops — the section is retried automatically • Files already in the target language aren't translated again (no API cost) • Filler-word cleanup now has custom options for what it removes • Better line breaks in formatted SRTs • Extra exports (formatted, TXT, VTT) can now be created from added SRT files

v2.6.6 — More Accurate, Faster Translations • Fixed: Long translations no longer fail or lose subtitles. Every line is checked, so subtitles stay complete and in sync. • Improved: Claude translates up to 5× faster, about 2 minutes per hour of video. • Improved: Gemini is quicker, and it switches to a backup model when the free daily limit runs out. • New: Formatted SRT styles: Streaming / web (42 × 2) and Broadcast UK (38 × 2), or Custom. • Improved: AI models update without updating the app.

Improved AI model update system and updated per job cost estimates.

Prompt enhancement for custom languages.

Updated AI models.

Fixed an issue with the in-app updater for non-app store versions.

What's new in Transcriber Translator 2.6.0 NEW - Filler-word & stutter cleanup — optional one-click export that creates a cleaned copy of your subtitles with fillers ("um", "uh", "sort of", "you know") and stutters ("I - I think" → "I think") removed. Non-destructive — your original SRT and translations are untouched. Runs locally, no extra API cost. Also cleans English translations of non-English files. - Plain-text transcript (.txt) export — a clean, timestamp-free transcript alongside every SRT, ideal for show notes, articles, and documents. - WebVTT (.vtt) export — .vtt subtitles alongside every SRT for web video players and HTML5 platforms. - Speaker labels — optional "[speaker_0]" prefixes in new transcriptions using ElevenLabs speaker detection; great for interviews and multi-person recordings. IMPROVED - Upgraded to Claude Sonnet 5 for translation — better, more nuanced quality at the same price. - Gemini translation now runs on Gemini 2.5 Flash with an automatic fallback for added reliability during busy periods. - More reliable translation of very long files — large files are split into smaller, individually-verified pieces with automatic retries, so lengthy webinars and films complete cleanly without dropped lines. - Pause now takes effect mid-file (between translation sections) instead of waiting for the whole file to finish. - Per-file source language is now correctly applied — setting a specific transcription language on an individual file now takes effect. - Clearer error messages — transcription errors now show the actual reason (e.g. quota exceeded, invalid key) instead of a generic failure. - More accurate usage tracking — local minutes now align with your ElevenLabs billing cycle. FIXED - Large files (up to 1GB) no longer cause a memory spike during upload. - Two queued files with the same name no longer show progress on the wrong row. - Progress bar no longer jumps backwards on mixed queues. - Stopping the queue no longer shows a "Queue Finished" message; all output files are revealed together in Finder when the queue finishes.