Killer whales (Orcinus orca) exhibit significant ecological and genetic diversity, with three primary sympatric ecotypes in the Northeast Pacific: Resident, Bigg’s (Transient), and Offshore. Each ecotype is adapted to distinct ecological niches, with unique foraging strategies, social structures, and vocal behaviors. These differences underscore the challenges in monitoring and conserving each group, as they coexist within overlapping geographic ranges yet do not intermix. This dataset, compiled from diverse sources, provides a comprehensive resource for the detection and classification of killer whale vocalizations. The >225,000 call-level annotations collected from 23 locations, and a cleaned annotation csv that collates annotations across the different protocols. Recordings spanning eleven years from various geographical locations within the Northeast Pacific Ocean, collected using multiple hydrophone systems. It addresses the challenge of differentiating killer whale calls from other marine species and environmental noise and includes specific instances of confounding signals to enhance model robustness. Detailed annotations capture a broad spectrum of vocalizations and associated metadata, facilitating the development of advanced machine learning models for ecological monitoring. This curated dataset aims to improve the accuracy of killer whale detection algorithms, support conservation efforts, and advance our understanding of killer whale acoustic communication across different ecotypes.