Voice assistant software struggles with minority dialects, Georgia Tech study finds

New study finds some voice assistant software may have difficulty transcribing English speakers with minority dialects. (Solen Feyissa and Reet Talreja)

Andy Wong / Andy Wong

A new Georgia Tech and Stanford study shows automatic speech recognition (ASR) models, used in voice assistants like Amazon Alexa, may not be as accurate when transcribing English speakers with a minority dialect.

However, the study found the transcription of Standard American English (SAE) “significantly outperformed” three dialects: Spanglish, Chicano English and African American Vernacular English.

On Thursday’s edition of “Closer Look,” Camille Harris, a Ph.D. candidate in computer science at Georgia Institute of Technology and lead author of the study, discussed some of the key findings. Harris also talked about the impact of code-switching on ASR technology and the need for more diverse representation in tech.