Attentional focus and emotion modulate voice recognition deficits in cerebellar stroke patients

The cerebellum, long regarded as a motor structure, is increasingly recognized for its role in higher-order cognitive and socio-emotional functions. Its contribution to vocal emotion decoding, however, remains insufficiently understood. While prior work has linked the cerebellum to attentional control and predictive coding, direct evidence for its role in modulating prosody recognition under explicit versus implicit attentional demands is lacking. This study investigated how cerebellar stroke—controlled for time since stroke in months, and with an equal number of left- and right-lateralized lesions—alters vocal emotion processing depending on attentional focus. We aimed to disentangle sensory encoding and integration from decisional contributions of the cerebellum by combining behavioral analyses, drift diffusion modeling (DDM), and functional MRI in cerebellar stroke. Fifteen patients with chronic, isolated cerebellar stroke and fifteen matched controls performed two tasks on vocal stimuli expressing anger, happiness, or neutrality. In the explicit task, participants categorized the expressed emotion; in the implicit task, they categorized speaker gender while ignoring emotional tone. Behavioral performance was analyzed using mixed-effects logistic regression and DDM (angle model). Functional MRI analyses included conventional contrasts as well as model-based regressors derived from behavioral and computational parameters. Patients showed lower accuracy than controls, with deficits particularly pronounced in the explicit task. Performance also varied by emotion: gender recognition for angry voices (implicit processing) was relatively preserved, whereas happy and neutral voices were more vulnerable. Contrary to predictions, DDM parameters did not differ between groups. Instead, task effects dominated: implicit processing yielded higher drift rates, larger boundary separation, faster non-decision times, and stronger attentional focus compared to explicit recognition, suggesting that explicit evaluation imposes additional cognitive costs. At the neural level, patients recruited extended networks during explicit processing, including orbitofrontal cortex, amygdala, anterior insula, and cerebellar lobule IX, alongside cerebello-cortical tracts. Implicit processing was associated with more restricted activations, particularly in frontal opercular and cerebellar regions (lobule IX). Model-based analyses further revealed that successful categorization in patients relied on the left inferior frontal gyrus, inferior parietal lobule, and pre-supplementary motor area, although these effects were not significant when controlling for time since stroke. Our findings demonstrate that the cerebellum does not primarily shape decision dynamics but optimizes sensory representations of prosodic cues for downstream evaluation. When predictive tuning is compromised, patients maintain intact decision processes yet rely on compensatory cortical recruitment, particularly during explicit tasks. This pattern supports predictive coding accounts of cerebellar function in socio-emotional communication and highlights the need to consider subtle socio-affective deficits in cerebellar patients.