HRTF Personalization from Anthropometrics: Trained Neural Network Models

Trained Keras model weights for predicting personalized Head-Related Transfer Functions (HRTFs) from anthropometric measurements (head, ear dimensions). Includes multiple architecture variants: stacked pipeline (personalization-1.0.0 through 1.0.2, 1.1.0, 1.1.1) and a single fully-connected network (only_magtotal_fully_connected). Each variant is trained on the CIPIC dataset with experiments across dropout rates, dataset variants (cipic, cipic-corr-height, Smoot smoothed variants), and loss masking strategies (lateral masking, left-right weighting, single-point/area focus). Archive contains kmodels/ (full Keras models), weights/ (best-checkpoint weights), mse/ and val_loss/ (training curves).