Proper singing requires the integration of auditory feedback mechanisms with the vocal motor system, such that vocal pitch can be precisely controlled. To determine the neural substrates involved in audio-vocal integration, non-musicians and experienced singers underwent fMRI scanning while they sang a single tone with either unaltered ("simple") or pitch-shifted auditory feedback; in pitch-shifted trials, subjects were instructed either to ignore or compensate for the shifted feedback. We hypothesized that the anterior cingulate cortex (ACC), superior temporal gyrus (STG), and anterior insula may be involved in audio-vocal integration due to their functional roles during singing and their anatomical connectivity. Although singers were more accurate than non-musicians in simple singing, both groups recruited similar functional networks. Singers ignored the shifted feedback better than non-musicians, and both groups also displayed different patterns of neural activity for this task: singers recruited bilateral auditory areas and left putamen, while non-musicians recruited the left supramarginal gyrus and primary motor cortex. While there were no significant group differences in performing the compensate task, singers displayed enhanced activity in the ACC, superior temporal sulcus, and putamen, whereas non-musicians exhibited increased activity in the dorsal premotor cortex, a region involved with sensorimotor interactions. We propose two neural substrates for audio-vocal integration: the dorsal premotor cortex may act as a basic interface, but with vocal training and practice, the ACC, auditory cortices, and putamen may be increasingly recruited as people learn to monitor their auditory feedback and adjust their vocal output accordingly.