Extended Data Fig. 4: Comparing DNN encoding performance across different convolutional layers in the HuBERT model for AN and IC neurons.
From: Dissecting neural computations in the human auditory pathway using deep neural networks for speech
a) The brain prediction score of the best-performing neural encoding model based on each single layer (the 4th – 7th CNN layers and the final convolution output) in the HuBERT model model (maximum over delay window length). b) The averaged brain prediction score at CNN4 – CNN7 in the HuBERT model with different delay window lengths. Note that the sampling rates vary at different layers: CNN4–400 Hz, CNN5 – 200 Hz, CNN6 – 100 Hz, CNN7 & CNN out – 50 Hz. AN: light shaded bars; IC: dark shaded bars. Box plot shows the first and third quantiles across electrodes, orange line indicates the median, gray line is the mean value, and whiskers indicate the 5th and 95th percentiles. * p < 0.05, ** p < 0.01, *** p < 0.001, two-sample t-test, two-sided, n = 50 unique neurons for AN, n = 100 unique neurons for IC.
