Modeling Paralinguistic Features for Speech Neuroprostheses: A Systematic Review

Recent advances in brain-computer interface (BCI) technology have brought speech neuroprostheses closer to restoring naturalistic communication for individuals with severe motor or speech impairments. Although current systems can decode phonemic and lexical information from cortical activity, they often fail to capture paralinguistic dimensions such as intonation, prosody, loudness, and affect that are essential for conveying emotion and intent. This review synthesizes emerging research on modeling and incorporating paralinguistic features into speech neuroprosthetic design. Foundational studies have mapped the cortical organization of speech production, identifying articulatory somatotopy within the sensorimotor cortex and functional subdivisions across planning, phonological, and semantic representations. More recent work has characterized expressive encoding within the ventral precentral and supplementary motor areas, while peripheral and multimodal sensing approaches have demonstrated promising accuracy in decoding emotional states. Collectively, these findings signal a shift toward closed-loop neuroprosthetic systems that integrate affective modulation with linguistic decoding, enabling restoration of speech that is not only intelligible but expressively human.