Virtual humans are often presented as mixed reality characters projected onto screens that are blended into a physical setting. Stereo loudspeakers to the left and right of the screen are typically used for virtual human audio. Unfortunately, stereo pairs can produce an effect known as precedence, which causes users standing close to a particular loudspeaker to perceive a collapse of the stereo sound to that singular loudspeaker. We studied if this effect might degrade the presentation of a virtual character, or if this would be prevented by the ventriloquism effect. Our results demonstrate that from viewing distances common to virtual human scenarios, a movement equivalent to a single stride can induce a stereo collapse, creating conflicting perceived locations of the virtual human's voice. Users also expressed a preference for a sound source collocated with the virtual human's mouth rather than a stereo pair. These results provide several design implications for virtual human display systems.