Privacy-Enhancing Infant Cry Classification with Federated Transformers and Denoising Regularization

TL;DR

This paper introduces a privacy-preserving infant cry classification system using federated transformers and denoising autoencoders, achieving high accuracy and efficiency suitable for real-time edge deployment.

Contribution

It presents a novel end-to-end pipeline combining federated learning, denoising autoencoders, and transformers for privacy-aware infant cry analysis.

Findings

Achieved macro F1 score of 0.938 and AUC of 0.962.

Reduced client upload from 36-42 MB to 3.3 MB per round.

Real-time inference at 96 ms per second of audio on NVIDIA Jetson Nano.

Abstract

Infant cry classification can aid early assessment of infant needs. However, deployment of such solutions is limited by privacy concerns around audio data, sensitivity to background noise, and domain shift across recording environments. We present an end-to-end infant cry analysis pipeline that integrates a denoising autoencoder (DAE), a convolutional tokenizer, and a Transformer encoder trained using communication-efficient federated learning (FL). The system performs on-device denoising, adaptive segmentation, post hoc calibration, and energy-based out-of-distribution (OOD) abstention. Federated training employs a regularized control variate update with 8-bit adapter deltas under secure aggregation. Using the Baby Chillanto and Donate-a-Cry datasets with ESC-50 noise overlays, the model achieves a macro F1 score of 0.938, an AUC of 0.962, and an Expected Calibration Error (ECE) of 0.032, while reducing per-round client upload from approximately 36 to 42 MB to 3.3 MB. Real-time edge inference on an NVIDIA Jetson Nano (4 GB, TensorRT FP16) achieves 96 ms per one-second spectrogram frame. These results demonstrate a practical path toward privacy-preserving, noise-robust, and communication-efficient infant cry classification suitable for federated deployment.