Ear-worn devices (earables) are increasingly central to smart system interactions involving privacy-sensitive data, yet secure user authentication on these devices remains a challenge. Existing methods often depend on auxiliary sensors like in-ear microphones or accelerometers, which are absent in many commercial earables. This paper introduces a novel biometric approach leveraging natural head gestures. We observe that head gestures generate unique air friction patterns detectable by out-ear microphones, producing sonic signatures shaped by the head and neck's musculoskeletal dynamics. These signatures serve as a robust basis for earable authentication. We propose HMPrint, a system that captures air-friction-induced sonic effects (AFiSe) from head gestures via outear microphones for authentication. HMPrint incorporates advanced spectral analysis, synthetic data generation using variational autoencoders, and a contrastive continual learning framework to enhance robustness against inconsistent wearing postures, varied movement patterns, and environmental noise. A proof-of-concept prototype was tested with 30 participants and 15 commercial earable models across diverse conditions. Results show that HMPrint achieves high authentication accuracy (97.49% recall), a low FAR (2.34%), and strong resistance to spoofing (98% success rate).