Researching the viability of Speech Imagery as a BCI paradigm

Speech Imagery (SI) is envisioned as a premier paradigm for Brain-Computer Interface (BCI) design. As an endogenous modality, it offers a highly intuitive control scheme—allowing users to command devices by mentally articulating words without the need for external sensory stimulation. While the field has gained significant momentum, resulting in a diverse body of literature, this thesis identifies a critical gap between experimental optimism and practical implementation. We aim to progress the consolidation of the SI paradigm by rigorously assessing its feasibility, reproducibility, and replicability. Our analysis of 104 decoding attempts revealed a striking disparity: only 6% were conducted in real-time scenarios, despite over half involving original data collection. To investigate this "reproducibility crisis," we attempted to reproduce findings from two popular SI datasets and compared them against the well-established Motor Imagery (MI) paradigm. Our results uncovered significant challenges; reproduction performance was consistently lower than originally reported, with 50% of attempts showing performance drops of up to 40% due to incomplete methodological reporting and flawed evaluation procedures. Furthermore, a large-scale replicability analysis across 12 heterogeneous SI datasets painted a challenging picture, with the majority of datasets failing to meet practical BCI efficiency thresholds. However, this comprehensive evaluation also identified a definitive path toward paradigm maturation. We discovered that rhythmic protocols in speech imagery lead to superior decoding results. Our meta-analysis suggests that rhythmic repetition may act as a organizational template, inducing more structured and consistent neural responses characterised by lower covariance matrix entropy—which significantly facilitates decoding. We conclude by proposing a set of rigorous directives for the field, including standardised reporting frameworks and rhythmic task designs, to transition Speech Imagery from an experimental curiosity to a consolidated and reliable BCI technology.