In conditional autonomous driving, the driver will return to manual driving in response to a takeover request from the system. Due to technical limitations, autonomous driving systems sometimes issue incorrect takeover request, which may affect the driver's human-machine trust and subsequent takeover decisions. In addition, different auditory warning types of takeover auditory types may also affect the driver's response to takeover. Therefore, it is important to explore the on human-machine trust and takeover performance under different takeover auditory types for the design of takeover cues for autonomous driving systems. It has important implications for the design of takeover request for autonomous driving systems. With pedestrian warning as an example, a simulated driving experiment was conducted to explore the effects of auditory types (speech/earcon) and takeover cue error types (miss/false) on driver's human-machine trust and driving behaviour, both in terms of performance and subjective evaluation. Drivers have shorter takeover reaction time and higher usability and satisfaction with earcon than speech. After experiencing miss request, drivers have lower levels of human-machine trust and faster response to subsequent correct takeover request.