@ltiatcmu.bsky.social '23 + @scsatcmu.bsky.social '22 + ex-Amazon. eng、華語、tâi-gú
We find script (e.g. Cyrillic, Latin) to be a linear direction in the activation space of Whisper, enabling transliteration at test-time by adding such script directions to the activations — producing e.g. Cyrillic Japanese transcriptions.
We find script (e.g. Cyrillic, Latin) to be a linear direction in the activation space of Whisper, enabling transliteration at test-time by adding such script directions to the activations — producing e.g. Cyrillic Japanese transcriptions.
#Interspeech2025
#Interspeech2025