MetrikaBox: An open framework for experimenting with audio classification

Abstract

This paper presents MetrikaBox, a general-purpose, open-source, and extensible audio classification package designed to facilitate the development of Deep Learning (DL) models for a wide range of audio processing tasks. The software manages all necessary preprocessing steps to build classification models capable of distinguishing between user-defined classes using advanced Artificial Intelligence (AI) techniques. MetrikaBox is well suited for tasks such as musical genre classification, voice-versus-music discrimination, and other audio classification or segmentation applications. Users can either employ the package as provided or extend it by integrating their own datasets, classification models, data loading systems, augmentation techniques, and more. The package has been tested in both commercial and academic settings, where it has produced models for industrial audio processing and served as a platform for proof-of-concept applications. Comprehensive documentation and practical examples included in the repository support users in integrating the system into their audio analysis projects. MetrikaBox is openly available and provides a user interface for convenient testing.