vak: a neural network framework for researchers studying animal acoustic communication

How is speech like birdsong? What do we mean when say an animal learns their vocalizations? Questions these are answered by studying how animals communicate with sound. As in many other fields, the study of acoustic communication being revolutionized deep neural network models. These models enable answering questions that were previously impossible to address, part because automate analysis very large datasets. Acoustic researchers have developed multiple for similar tasks, often implemented as research code one several libraries, such Keras and Pytorch. This situation has created a real need framework allows easily benchmark models, test new own data. To address this need, vak (https://github.com/vocalpy/vak), designed researchers. (\textquotedbl{}vak\textquotedbl{} pronounced \textquotedbl{}talk\textquotedbl{} or \textquotedbl{}squawk\textquotedbl{} was chosen its similarity Latin root voc, \textquotedbl{}vocal\textquotedbl{}.) Here describe design vak, explain makes it easy apply We highlight enhancements made version 1.0 significantly improve user experience library. provide without expertise learning access can be run via command-line interface uses configuration files. Vak also used directly scripts scientist-coders. achieve this, adapts patterns API from domain-specific PyTorch libraries torchvision, modules representing operations, datasets, transformations pre- post-processing. leverages Lightning library backend, so developers users focus on domain. proof-of-concept results showing compare existing model families. In closing discuss our roadmap development vision community users.