Deep below Morrison Hall, quadraphonic artist Suzanne Ciani sat in a soundproof recording studio, reciting strings of non-sequiturs with varying intonations to form a deepfake program.
Researchers and engineers used Ciani’s non-sequiturs to generate training data for a text-to-speech program, or deepfake, that will pre-record messages that Ciani fans can purchase on vinyl records.
“The lazy cow is resting in the fresh grass,” Ciani said in an excited voice during the recording. “Jump over the fence and dive in.” The freezing air passed through the mantle. Add quick leads to wrong sums.
Justin Norman, a doctoral student at UC Berkeley’s School for Information, said Ciani’s speech synthesizer would be capable of generating messages never recorded before. Other programs, including Apple’s Siri, use artificial intelligence, or AI, which samples information from a certain voice.
Norman added that the speech synthesizer will have attributes that will make it more realistic, including emotion, intonation, continuous flow, and normal speaking cadence.
Rafael Valle, who has an interdisciplinary doctorate from UC Berkeley, recorded Ciani on Friday. He said the synthesizer will also have the ability to modulate emphasis, tone and pitch.
Valle added that the synthesizer can potentially recite messages in any language, with sufficient sophistication.
“Read all questions with enthusiasm or joy – the synthesizer has the ability to read the emotion, not the content,” Valle told Ciani during the recording session. “You are already laughing, it’s great. It brings us as close as possible to a real voice. Laugh as much as you want.
The producer of the KamranV project noted that Ciani’s “authentic” messages, which she never really said, are a tribute to Don Buchla.
Buchla, who died in 2016, invented an electronic musical instrument called a modular synthesizer, studied physics on campus, and employed Ciani after earning her Masters in Composition at UC Berkeley.
“It’s like an autograph,” Ciani said. “It’s a personalization… You have the message and the recipient of the message and the content is created in that interchange – they create the message when they receive it. “
Ciani added that she would never allow anyone to use the text-to-speech program to generate messages that she did not approve of beforehand. The goal of the project is not to generate disinformation, she added, but rather to play with new technology and communicate directly and indirectly.
Recent uses of deepfake technology are making Ciani ‘nervous’. She said she felt “threatened” that the line between real and automated had become much blurred. “Alternate realities,” such as movies, already exist, she noted, but the difference between movies and some deepfake technology is that people know movies are fake.
“These things are starting to take over our real lives and people are having a hard time dealing with it,” Ciani said. “We have to be able to bring the doubt or the skepticism, the analysis, the assessment of where they come from.”
Contact Eric Rogers at [email protected], and follow him on Twitter at @eric_rogers_dc.