Introduction

audformat was created to handle all databases inside audEERING in an unified way to make them easily accessible.

The main challenge in specifying the format was to make the format universal enough to be applicable to many use cases as possible, but simple enough to be understood and parsed easily.

audformat can store the following information:

• Raters (humans, aggregated, machine)

• Labels (categorical, numeric, etc.)

• Annotations (file or segment, rater, confidence, multi-label)

• Metadata (recording date/location, source, speaker age/gender, etc. and may apply to whole corpus or individual files / segments)

The format further allows to

• link audio files to meta data and annotations

• use generic tools to access the data, create statistics, merge annotations

• search / filter information in databases

• use machine learning tools to easily access data and run experiments across databases