Self-Supervised Pretext Tasks for Infant Cry Analysis: A Controlled Comparison and a Cautionary Result on Donateacry
2026-08-31 • Machine Learning
Machine Learning
AI summaryⓘ
The authors tested six different ways a computer could learn from audio without labels to recognize and understand baby cries, all using the same small model and data. They found that detecting cries works well with certain methods, even if the model never heard a cry before. However, figuring out why a baby is crying was no better than guessing, and increasing model size or tweaking data didn’t help. They showed past studies claiming high accuracy likely had data mistakes that made the task easier. Their work suggests the real challenge lies in the labels and data quality, not the models themselves.
self-supervised learninginfant cry analysisencoded modelscry detectioncry-reason classificationlinear probemasked spectrogramdata leakageaugmentationevaluation protocol
Authors
Luigi Simeone
Abstract
We compare six self-supervised pretext tasks for infant cry analysis under a fixed budget, meaning the same compact encoder of 1.17M parameters, the same 115 hours of license-verified public pretraining audio, and the same evaluation protocol for every candidate. On cry detection the reconstructive objectives dominate, and a linear probe over a masked-spectrogram encoder reaches 0.988 AUC with subject-wise splits even though the encoder never observed a cry during pretraining. On cry-reason classification over donateacry, the de facto public benchmark for cry reasons, every encoder performs at chance (0.38 to 0.54 macro AUC over 5 classes), and neither domain adaptation on 1.8 hours of real cries nor end-to-end fine-tuning moves the result. Since a frozen HuBERT-base with 80 times more parameters shows the same pattern, the bottleneck must sit in the labels and not in model capacity. We then reproduce the 90\%+ accuracies of the donateacry literature on our own system by changing nothing but the evaluation protocol: clip-wise splits raise accuracy to 85.2% (barely above the 83.8% majority-class baseline), and applying augmentation before splitting raises it to 97.9%, matching the reported state of the art, from the same model that measures 0.49 macro AUC under subject-wise splits. Under leakage-free splits, a twentyfold augmentation of the labeled set (vocoder speaker perturbation and noise mixing, 21 hours) leaves cross-subject AUC unchanged: for this task the effective sample size is the number of infants. We release code, seeds and per-clip license manifests.