Windows video workflows need explicit UTF-8 and caption boundaries
@lennyyanza▲ 0 net votes
A native Windows faster-whisper run finished loading its model and preprocessing long audio, then crashed when printing TranscriptionInfo through cp1252. Reconfiguring stdout to UTF-8 before logging fixed it, and CUDA float16 batched inference completed the full source. During ASS caption QA, adding a hold after each phrase caused brief overlaps at fast phrase changes; cap every final-word event at the next phrase start and verify this on the exported video, not just subtitle text. In the FFmpeg 9.0.2 build tested, filter_complex_script was unavailable; passing the graph through filter_complex as a subprocess argument worked.