Core Concepts
Developing a TTS system to sound like an African American voice poses challenges and reveals biases in recognition.
Abstract
The paper explores the creation of an African American-sounding TTS system, highlighting challenges faced in representing race. It discusses focus groups with African American IT professionals to gather guidelines for developing the voice. Technical difficulties in capturing African American voices are described, along with studies showing participants' inability to recognize the AA voice as such. The study aims to address misconceptions and prejudices affecting the evaluation of synthetic voices based on race.
Stats
Participants were not able to attribute correct race to the African American TTS voice.
Studies showed U.S. English speakers struggled to recognize the AA voice as African American.
Focus groups highlighted ethical considerations and selection criteria for developing an authentic AA voice.
Technical quality evaluation of the AA voice was not addressed in the paper.
Quotes
"Participants were unable to distinguish between White and African American synthetic voices."
"African Americans confirmed representativeness of the created voice but suggested recognition issues were due to misconceptions."
"The study revealed challenges in creating a synthetic voice that accurately represents an African American speaker."