How did speech evolve? This and other questions answered by the study of animal acoustic communication require cutting edge computational methods, big team science, and true interdisciplinary collaboration. To those ends, more and more research groups are sharing their code. However, this code is often written to answer specific questions and tailored to specific data formats, which vary widely across groups. As a result, it is often difficult to read and reuse shared code. Furthermore, because each group writes code to deal with low-level details, there is massive duplication of effort. Here I present a software community, VocalPy, created to address these issues. Its core goal is to provide an interoperability layer. It achieves this with high-level abstractions, e.g., for audio, spectrograms, and annotation formats. These abstractions allow researchers to write in idioms designed to be readable across groups, while moving details about formats to configuration options. I describe how as a proof-of-concept we are leveraging VocalPy packages to develop a framework for benchmarking neural network models named vak. VocalPy enables vak to work with many data formats when preparing datasets for model training or automated analysis. In closing I outline a roadmap for future work.