A fixed test did not confirm shared voice interaction gains in 20 iEEG patients and 16 fMRI adults

We tested whether human voice class changes the acoustic slopes needed to predict cortical sound responses. Separate public studies supplied intracranial local field potentials and fMRI. Four people per modality supplied development. The fixed test used 20 other patients and 16 other adults. Six outer sound folds and three inner folds separated response tests from model fits and penalty choice. The baseline had linear acoustic terms and a voice mean shift. The test model added voice-by-acoustic terms. Mean gains against intercept error were -1.922 percentage points in iEEG and +1.002 in fMRI. Their 95% person Student intervals were -4.439 to +0.594 and +0.468 to +1.537 points. Gains were positive in 10/20 and 15/16 people. The fixed shared rule failed; conjunction sign p was 0.58810. The fMRI gain became uncertain when a fixed-prediction resample also varied sounds. Each separate fMRI response estimate had a negative mean gain. A later check kept the same linear function space but changed ridge coding. It changed the voice-specific fMRI result. A smooth acoustic-only model also produced a large interaction gain. Thus, a gain cannot identify a category cause. These are new endpoint tests in reused studies. Cohort, sound, coverage and signal differences limit scope. The package gives code, derived electrical inputs, all scores, source checks, plans, figures and repeat steps. No positive cross-modal generalization is claimed.

auditory cortex · voice · intracranial EEG · fMRI · acoustic prediction · ridge regression · interaction · cross-validation · negative result