USVexplorer: Robust Detection of Ultrasonic Vocalizations with Cross Species Generalization

Ultrasonic vocalizations (USVs) emitted by animals carry rich emotional, social, and behavioral information beyond human hearing range. While existing USV detection methods have made progress, they suffer from limited cross-species generalization and typically fail to align acoustic signals with behavioral data, limiting understanding of sound-behavior relationships. We propose USVexplorer, an end-to-end framework that integrates 1D convolutional layers with a Transformer encoder for robust USV detection. USVexplorer significantly outperforms state-of-the-art methods on rat datasets, while demonstrating strong cross-species generalization. Additionally, USVexplorer provides audio-video synchronization capabilities, enabling multimodal analysis for animal behavior research.