Schedule 71. StuTS + 31. TaCoS

Previous research has touched upon bias in automatic speech recognition (ASR) but has mostly focused on the English language. However, as these systems are used around the globe, finding biases in different languages is of high importance. The current study seeks to find biases based on age, region, and sex in a Dutch state-of-the-art ASR model. The Dutch data from the Spoken Dutch Corpus will be used to train the hybrid DNN-HMM model and it will be evaluated using the Flemish data from the JASMIN-CGN corpus. In addition, I will provide an in-depth phoneme analysis. The findings will contribute to a better understanding of bias, and subsequently inclusivity, in ASR.