Abstract The ability to read out, or decode, mental content from brain activity has significant practical and scientific implications 1 . For example, technology that translates cortical activity into speech would be transformative for people unable to communicate as a result of neurological impairment 2,3,4 . Decoding speech from neural activity is challenging because speaking requires extremely precise and dynamic control of multiple vocal tract articulators on the order of milliseconds. Here, we designed a neural decoder that explicitly leverages the continuous kinematic and sound representations encoded in cortical activity 5,6 to generate fluent and intelligible speech. A recurrent neural network first decoded vocal tract physiological signals from direct cortical recordings, and then transformed them to acoustic speech output. Robust decoding performance was achieved with as little as 25 minutes of training data. Naïve listeners were able to accurately identify these decoded sentences. Additionally, speech decoding was not only effective for audibly produced speech, but also when participants silently mimed speech. These results advance the development of speech neuroprosthetic technology to restore spoken communication in patients with disabling neurological disorders.