At one fixed penalty, slope labels gave mean gains below 1 percentage point in 20 iEEG patients and 24 fMRI adults

We tested whether rise and fall slope labels improve prediction of human auditory responses. Four acoustic pairs measured pitch and loudness slopes. The reduced input kept both values of each pair and removed their labels. The other 80 descriptors remained. The augmented input also kept the four signed differences. Five folds tested sounds excluded from fitting. Separate studies supplied 20 reserved iEEG patients, 16 B2021 fMRI adults and eight NH2015 fMRI adults. Mean added gains were -0.557, -0.163 and -0.174 percentage points of class-reference error. Their 95% person Student intervals were [-0.904,-0.210], [-0.296,-0.031] and [-0.532,+0.184] points. The shared positive rule failed. A fixed test placed every mean below one percentage point, with Holm p=0.0001104 across two claims. This bound assumes independent people and adequate Student inference for person scores. It concerns one fixed ridge penalty and selected sound panels. Primary electrical total skill was uncertain. A planned stronger penalty gave positive total skills and small label gains in all studies. Actual reversal of 165 waveforms did not exchange the slope descriptors exactly. The four correlations with negative reversed differences ranged from 0.377 to 0.811. Thus, this input test does not isolate neural time reversal. The contribution is a controlled benchmark, a bounded negative result in both modalities, and a forward/reversed acoustic dataset. Separate cohorts, different sounds, an unknown electrical feature build and source processing limit scope. No common neural cause is established. Code, permitted inputs, source pins, all results and repeat instructions are preserved.

auditory cortex · intracranial EEG · fMRI · acoustic features · rise and fall slopes · time reversal · prediction limits · ridge regression