Replace HTDemucs with BS-RoFormer for superior vocal separation quality

Wrap voice mapping in QScrollArea to prevent UI overflow with many speakers

Context-padding for short sentences: pad with filler, infer filler alone, trim by duration difference

Fix TTS crash on short sentences: merge short ones and pad with ellipsis

Fix aggressive tail trim: increase end margin from 5ms to 80ms

Approach 3: incorporate gap into stretch calculation instead of shifting audio

Add min gap between dubbed audio clips to prevent overlap

Auto-run Demucs in CapCut export if vocal/inst files missing

Auto-detect vocal/instrumental files on disk for CapCut export

Add stretch threshold setting in CapCut export dialog (default 0.3s)

Smart time-stretch: split video at long audio segments and slow down to match TTS duration

Add WAV post-processing: trim silence and remove initial noise artifacts

Fix instrumental missing environmental noise by using phase subtraction

Export vocal and instrumental separated tracks to CapCut

Disable mouse wheel scrolling on Comboboxes

Revert back to Pyannote 3.1 but keep Demucs preprocessing

Replace pyannote with SpeechBrain ECAPA-TDNN + Spectral Clustering

Add Demucs vocal separation for better diarization

Update core source code

Add files using upload-large-folder tool