To use different audio data sets in machine learning pipelines across our teams at audEERING we developed audformat as a common format that supports labeled audio or video data. On top of the format audb adds the possibility to version and share the data easily.

Scientific findings depend to a large degree on software and data. Consequently, scientific publications should evolve over time as they get tested, improved, or even falsified. To implement solutions that reach beyond the classical journal paper, I tried to present them as HTML pages, generated from a corresponding git repository and data publications. I selected sphinx-doc instead of Jupyter notebooks for implementation as it allows for a more modular approach and re-usability.

The study of sound field synthesis techniques involves numerical simulations, which are crucial to evaluate them. In an effort started at the Technical Universität Berlin we started developing and publishing the needed software. First implemented in Matlab/Octave and later in Python.

The Two!Ears EU project investigated computational modelling of how humans assign meaning to auditory scenes. I worked on how humans judge the sound quality of spatial audio scenes, reproducibility of results, and the management and versioning of software and databases.

The AABBA project is an informal group of researches working in the field of binaural auditory modelling. To foster reproducibility and better cooperation, we developed the SOFA format for exchange and storage of head related impulse responses. In addition, we collect implementations of models in the Auditory Modelling Toolbox.

Before switching to Python I used Gnuplot for producing plots for papers and web pages. I published a collection of howtos and converted common color pallettes to Gnuplot.