Bajaj, BettinaBoayrid, Nora2024-11-262024-09APA 7th editionNAhttps://hdl.handle.net/20.500.14154/73791Following the recent applications of ASR systems in the medical field, Miner et al. (2020) called for further research to examine the accuracy of medical ASR systems against different variables before widespread use. Among other variables, gender plays a prominent role in influencing the performance of ASR systems (Fucci et al., 2023). Hence, this thesis aims to fill in the gap in the literature and examine the robustness of the Amazon Transcribe ASR system against gender bias in the medical domain for the Arabic language. The thesis includes transcribing 17 episodes from the first season of the Medical TV Show broadcasted on AL-Arabi Kuwait TV. The transcription outputs include 20,000 words spoken by adult Arab male and female medical professionals (i.e. 10,000 words spoken by each gender group). The transcription outputs were evaluated using the Word Error Rate (WER) evaluation metric and the Mann-Whitney U statistical test. The findings show that the tool recognized the female speakers better than the male speakers. The WERs for the female and male groups are 7.04 % and 13.03 %, respectively. While the females’ voice qualities influenced their WERs positively, the differences in the WERs between the male and female groups are not statistically significant (i.e. p-value > 0.05). In fact, some errors were found in the transcription outputs of both gender groups due to the tool’s architecture, the domain of the data, and the language in question rather than the speaker’s gender.89enAutomatic Speech RecognitionWord Error RateGender BiasArabic LanguageMedical DomainASR SystemsThesis