Table A1 Examples of decoded outputs mapped to voicing categories for selected minimal pairs

Target pair	Mapped as voiceless	Mapped as voiced
pan/ban	pan, pat, pain	ban, band, better, bender
tan/Dan	tan, ten, 10, tender, time	Dan, down, dandelion
cam/GAM	cam, KM, camp, can	GAM, game, gambler

The table illustrates representative Whisper outputs that were assigned to voiceless and voiced categories based on initial consonant identity after normalization.