With the development of speech synthesis technology, in-vehicle information systems are able to provide various voice personas for information prompts, meeting drivers’ personalized preferences and improving interaction experience. However, existing studies lack a systematic understanding of how different voice personas affect driver behavior under high cognitive load and non-driving task conditions. To examine the effects of different voice personas used for hazard warnings on drivers’ situation awareness while performing non-driving tasks with different immersion levels. A 2×2 experimental design was adopted, with two voice personas (dominant/personal assistant) and two immersion levels (high/low) as independent variables. Eighteen participants were recruited to perform a simulated driving experiment. In a video-based automated driving scenario, drivers’ situation awareness was evaluated by reaction time, total fixation duration on areas of interest, and questionnaire scores. Data analysis shows that compared with the personal assistant voice, the dominant voice significantly reduces drivers’ reaction time (P=0.006<0.05) and total fixation duration on area of interest (P=0.043<0.05), while achieving higher situation awareness questionnaire scores (P=0.029<0.05) under both immersion conditions. In potential hazard scenarios, hazard prompts delivered with a dominant voice persona help drivers detect dangers quickly and accurately interpret current driving behavior, thus enhancing their situation awareness during driving.