Speech perception and production are deeply interconnected, yet the cellular mechanisms that link them are unknown. Here, we recorded large-scale single-neuron activity from the human brain while participants performed a delayed sentence repetition task, which included listening to a sentence, waiting a short delay, and then speaking. Although the precentral gyrus (PrCG) is traditionally thought to be specific for motor control, we observed neuronal responses in single cortical columns that tiled across listening, delay, initiation, and speaking phases. In contrast, neurons in the posterior Broca’s area mainly distinguish the speaking phase and carry minimal speech information. Neurons in the superior temporal gyrus responded primarily to auditory input and feedback. Among the precentral neurons, we discovered a unique subset of neuronal responses that link perception and production, including “mirror” neurons that are transiently activated after hearing specific speech sounds and before speaking the same sounds. Another group called “bridge” neurons maintain elevated firing between the same speech elements from listening to speaking. Unsupervised population analysis reveals phasic latent components for each stage of the task, as well as sustained dynamics for specific sentences only in middle PrCG. PrCG neurons preferentially encoded higher-order representations of phonemic sequences, rather than simple articulatory or acoustic features. These findings demonstrate the complex, integrative role of the PrCG neurons in linking the perceived, produced, and mentally held speech via congruent speech sequence representations.