Speech-based Age and Gender Prediction with Transformers

F Burkhardt, J Wagner, H Wierstorf, F Eyben, BW Schuller, "Speech-based Age and Gender Prediction with Transformers," in 15th ITG Conference on Speech Communication, (2023). [ link ] [ paper ]

Bibtex

@proceedings{Burkhardt2023a,
    title     = {Speech-based Age and Gender Prediction with Transformers},
    author    = {Burkhardt, Felix and Wagner, Johannes and Wierstorf, Hagen
                 and Eyben, Florian and Schuller, Björn W.},
    booktitle = {15th ITG Conference on Speech Communication},
    publisher = {VDE},
    address   = {Aachen, Germany},
    month     = {September},
    year      = {2023}
}

Abstract

We report on the curation of several publicly available datasets for age and gender prediction. Furthermore, we present experiments to predict age and gender with models based on a pre-trained wav2vec 2.0. Depending on the dataset, we achieve an MAE between 7.1 years and 10.8 years for age, and at least 91.1% ACC for gender (female, male, child). Compared to a modelling approach built on handcrafted features, our proposed system shows an improvement of 9% UAR for age and 4% UAR for gender. To make our findings reproducible, we release the best performing model to the community as well as the sample lists of the data splits.

Supplementary material

ONNX versions of the 24-layer and 6-layer model are available at https://doi.org/10.5281/zenodo.7761387, together with the defined test splits from the paper. For an introduction how to use it, take a look at How to use our public age and gender model.