The irrelevant speech effect refers to the impairment of visual-information processing by background speech. Unlike the general noise, the studies on irrelevant speech effect mainly investigate the properties of speech on cognitive tasks. Most previous studies on irrelevant speech effect has focused on short-term memory for visually-presented word lists. They favored that the irrelevant speech effect is determined by the same process rather than the confusion between the contents of vocal and visual information. More concretely, the interference-by-process account considers the disruption as a result of conflict between deliberate processing in focal task and automatic processing of speech. However, little is known about the mechanism underlying the irrelevant speech effect in complex cognitive tasks. The present study extends this work by using measures of eye movements to examine whether the interference-by-process account also applies to the reading comprehension of Chinese sentences. Eye tracking provides a means for objectively recording and analyzing reading behavior in the normal environment, and its varied eye movement measures also permit examining subtle changes that occur during reading with background speech. 36 participants read sentences normally or being exposed to different types of background sounds. A 3 (background speech type: meaningful speech, meaningless speech, silence)×2 (task instruction: normal reading, proofreading) mixed design was used. The background speech type was the within-subjects factor, and the task instruction was the between-subjects factor. Processing in proofreading task is different from it is in normal reading. During proofreading, in addition to understanding the meaning of the sentences, the readers should pay more attention to the word-level information. The basic hypothesis was that if the irrelevant speech effect was due to the confusion of content, then we would expect that the interference occurred regardless of task instructions as the similarity of content between speech and sentence materials was not changed. Instead, if the interference was determined by the same process, we would predict an interaction between background type and task instruction. The accuracy rates of reading comprehension were not significantly different in all treatments. Linear-mixed effects models were used to analyze the eye movement data. The findings showed that: (1) Be consistent with many previous studies, the irrelevant speech effect on Chinese sentence reading was observed. The meaningful background speech reduced the reading rate, and leading the participants to make more fixations and regressions. While the meaningless speech had little influence on reading comprehension, which suggested that the semantic content of speech plays a more important role in the irrelevant speech effect on reading. (2) Furthermore, there was a significant interaction between background type and task instruction for reading rate and average fixation count per character. Specifically, the meaningful speech disrupted normal reading, while did not affect proofreading significantly. As the similarity between the contents of vocal and visual materials was identical, this interaction implied that it is the same process rather than the similar content determines the interruption. The present study supports the interference-by-process account. Some implications about the irrelevant speech effect in complex cognitive tasks were also discussed.