Models fit to voice sounds did not pass a shared animal sound test in 20 iEEG patients and 16 fMRI adults

We tested whether acoustic tuning fit only to voice sound classes predicts response differences among animal sounds. Separate human studies supplied signed intracranial EEG and BOLD fMRI. Four people per study supplied development. A fixed ridge model then scored 20 other patients and 16 other adults. No animal sound entered fitting. Mean within-animal response correlation was 0.03979 in iEEG, with a 95% person interval of -0.01039 to 0.08997. The fMRI mean was 0.23447, with interval 0.20215 to 0.26679. Common animal-label tests gave p=0.1952 and p=0.0002. The fixed shared rule failed. Mean centered R2 was -7.94689 and -0.05489. All 20 electrical models and 13 of 16 BOLD models had worse centered squared error than a constant. One electrical sound had a supplied descriptor 200.8 voice-training SD from the mean. Omitting that sound reduced error, but mean centered R2 stayed negative. Stronger shrinkage improved both study means; it remains a descriptive check. The BOLD voice class includes one synthetic speech clip. Refitting over sound samples left BOLD error loss uncertain. We retain all failures and shared-sound uncertainty. The result limits this model under the tested panels. It does not establish absent acoustic coding, a causal modality difference or a common neural source. Different cohorts, labels, feature sets, source selection and measurement filters limit scope. The package preserves code, permitted inputs, source terms and hashes, fixed plans, all results, figures and repeat instructions.

auditory cortex · intracranial EEG · fMRI · acoustic encoding · animal sounds · human voice · domain transfer · prediction calibration · negative result