Acoustic Interactions Vocal Repertoire and Mimicry
Using Deep Learning Techniques for Repertoire Analysis
Led by our research partner Peter Sugarman from Acoustic Interactions, this project investigates dolphin communication through comprehensive analysis of vocal behavior and mimicry capabilities. We analyze large datasets of dolphin vocalizations to extract whistle contours and categorize them into distinct clusters based on contour shape, distinguishing between signature and non-signature whistles using DeepAcoustics. Our research involves generating non-signature whistles that are unique within each group’s vocal repertoire. These custom whistles are then used to experimentally test dolphins’ vocal mimicry abilities, ultimately advancing our understanding of dolphin acoustic communication and vocal learning. The study integrates bioacoustics analysis, deep learning detection and classification techniques, and experimental behavioral testing to provide new insights into a critical dimension of dolphins’ complex communication systems.
More About DeepAcoustics in Our JASA Publication
Whistle Classification
DeepAcoustics enables automated clustering of dolphin whistles using custom feature weights such as start frequency, slope, and concavity of contour shape. Users can fine-tune clustering inputs and evaluate model performance with built-in tools like silhouette scoring to identify distinct vocal groups. This approach supports detailed analysis of whistle repertoires and vocal mimicry in dolphin communication research.
Explore Details in Sugarman's SMM 2022 PosterConsulting & Training Services
Technology trained consultants and partners are available help you with your project or training needs.
