In [PR 372](https://github.com/audeering/audb/pull/372#discussion_r1551821412), it was discussed that using [polars](https://pola.rs/) as a dataframe engine might be worth exploring in terms of its performance. Implementing this feature should add it to the [comparison benchmarks](https://github.com/audeering/audb/blob/dev/benchmarks/README.md).