Transfer learning from the state-of-the-art Whisper automatic speech recognition model can effectively predict the distribution of lexical responses perceived by human listeners to noisy speech stimuli, outperforming baseline methods.
