• New optional on-device speaker detection for up to eight speakers, with a compact model download of about 107 MB. • Listen to original cue audio and confirm speaker assignments; manually confirmed assignments are preserved. • Improved handling of brief speaker changes, overlapping speech, and single-channel speaker assignments. • Projects now save before audio generation and preserve pending edits when switching media or projects. • Safer regenerated-audio storage protects previously saved results if a save is interrupted. • Improved cancellation and error handling during Qwen recognition and speaker detection. • Additional stability and interface improvements.

Fixed repeated login Keychain permission prompts when checking trial and purchase status. Lifetime purchase verification continues through StoreKit. Includes additional stability improvements.

Version 2.0 expands the app into a complete local video translation and dubbing workflow: • Qwen3-ASR recognition with word-level timing • System voices plus four downloadable local AI voice models • Import, save, rename, reorder, and audition reference voices • Single- and multi-speaker dubbing with per-speaker parameters • Batch generation, per-cue playback, and regeneration • Duration overrun detection • Single-track, minimum-overlap multitrack, and professional editing project exports • Automatic saving to the app’s default output folder • Translation workflow and stability improvements