Introduction: Drumbeat.AI is a deep learning classification algorithm for analyzing otoscopic images, primarily trained on images from Indigenous children in Australia. Given the high risk of middle ear disease among Māori and Pasifika populations in New Zealand (NZ), this study aims to assess the algorithm's generalizability in a novel population of NZ ears. Methods: Otoscopic images, tympanometry, and audiometry data from urban NZ children aged 3-12 were collected at an audiology clinic. Three otolaryngologists labeled the images into four categories: Normal, Acute Otitis Media (AOM), Middle Ear Effusion (MEE), and perforation. Images lacking consensus or showing tympanic membrane retraction were discarded. The remaining images were split into training (100) and testing (50) datasets. The AI was retrained using the training set, and diagnostic performance was evaluated for accuracy, sensitivity, specificity, and area under the curve (AUC). Results: Out of 200 total datasets, 150 images were retained after exclusions. The model achieved an overall accuracy of 82%. After retraining with 100 NZ images, binary classification accuracy (normal vs. abnormal) improved to 84%, and disease-specific accuracy (normal vs. MEE vs. perforation) reached 82%. Initial testing had 11 misclassifications (7 as MEE), reduced to 9 after retraining (5 as MEE). Misclassifications of normal ears as MEE were more common in NZ children of European descent. Conclusions: The Drumbeat.AI model demonstrated improved diagnostic performance after retraining local NZ images. While overall accuracy increased, specific classes showed variable improvement: the classification of MEE and perforations improved significantly, whereas misclassifications for normal ears remained a concern. Future research with larger NZ-based training datasets would enhance the model’s performance and address the observed misclassifications.