Gender-neutral AI voice assistant developed to ‘combat gender bias’
Publish | 12 Mar 2019, 20:13
Barking orders at a digital device that responds in a woman's voice can reinforce sexist stereotypes, according to academics and creatives who launched the first gender-neutral artificial intelligence voice Monday.
Responding to such concerns, a Denmark-based team has created a voice nicknamed Q that was presented at the South by Southwest (SXSW) creative festival in Texas and is designed to be perceived as neither male or female.
"There is no reason that a voice has to be gendered," said Julie Carpenter, a research fellow in the ethics and emerging sciences group at California State Polytechnic University, who advised the project Q team.
"Emerging technology is being designed to rely on these ancient stereotypes."
Leading digital assistants such as Apple's Siri, the Amazon Alexa and Microsoft's Cortana are generally presented as female, Carpenter told Reuters.
All three have feminine names and mostly offer a woman's voice as the default setting, although Siri is set up to sound male in some languages.
"People seem to have a preference for female voices when the role of the AI is more supportive and to assist or help someone, while they associate male voices with an authoritative tone or an area of expertise," Carpenter said.
Apple and Amazon were not available for comment.
A spokeswoman for Microsoft said they had researched voice options for Cortana and found "a female voice best supports our goal of creating a digital assistant." She said the company had explored adding a male voice option to Cortana.
A team at Vice Media's Virtue creative agency came up with the concept for the Q voice to highlight concerns over gendered technology and offer an alternative, in a project for Copenhagen Pride.
"It isn't easy to create a genderless voice," said Nis Norgaard, a sound designer at Thirty Sounds Good studio, who produced Q.
He used research which found male voices are usually pitched between 85 to 180 hertz (Hz) and women's are typically between 140 to 255 Hz to identify a potential neutral range where the two overlapped.
It was not just pitch that defined the perceived gender of a voice — men tend to have a "flatter" speech style that varies in pitch less and they also pronounce the letters "s" and "t" more abruptly, said Norgaard.
The team working on Q recorded the voices of 22 transgender and non-binary people as a basis for the voice, both in an effort to represent a wider spectrum of gender and because it was thought they would sound less obviously male or female.
One final recording was chosen as Q, which was then digitally manipulated to make it more gender-neutral.
The finished voice was tested by more than 4,000 volunteers.
About half said they could not tell the gender and the other half were roughly evenly divided between guessing it was male or female.
Currently, users can only interact with Q via a website.
However, Emil Asmussen, a director at the Virtue agency, said he hoped it would push major tech firms to consider widening their digital AI options to include genderless voices.
"We really have high hopes this can create some debate around the world and start a dialogue going with some of the big tech companies," he said.
"They have a giant opportunity because we are on the brink of an AI revolution."
source: voa