User:Pbroskoff/Speech recognition#cite ref-:0 2-0
{{Dashboard.wikiedu.org draft template/about this sandbox}}
= Bias =
Speech recognition tends to understand certain races and genders better than others. Speech recognition has a worse performance for women and non-white English speakers when looking at accuracy in English. {{Cite journal |last=Martin |first=Joshua |last2=Wright |first2=Kelly |date=2022-12-14 |title=Bias in Automatic Speech Recognition: The Case of African American Language |url=https://academic.oup.com/applij/article/44/4/613/6901317 |journal=Applied Linguistics |volume=44 |issue=4 |pages=613-630 |via=Oxford Academic}}Speech recognition has difficulty with certain accents and dialects. The rate of accuracy depends on the dialect that is spoken. Scottish and Indian accents tended to be misunderstood by speech recognition programs, like YouTube's auto-captioning. {{Cite news |last=Bajorek |first=Joan Palmiter |date=2019-05-10 |title=Voice Recognition Still Has Significant Race and Gender Biases |work=Harvard Business Review |url=https://hbr.org/2019/05/voice-recognition-still-has-significant-race-and-gender-biases |access-date=2023-10-13 |issn=0017-8012}}
These biases exist because speech recognition software has had more exposure to certain types of voices, which can lead to the programs getting voices not included in the training mistaken. Companies that use speech recognition are taking steps to try and fix the bias, but the bias exists on many speech recognition platforms today.