Synthesia gives a journalist a talking likeness bound to one story
Synthesia made an interactive likeness of TechCrunch reporter Dominic-Madori Davis after photographing her and recording a short voice sample at its New York office. It answers questions about one of her investigations and redirects other questions to that story. The trial shows how a familiar face and voice can front a narrowly bounded conversation, while Davis's family found that the likeness could not answer personal questions.
Artificial Intelligence··Evening
Davis supplied photographs and voice
TechCrunch reporter Dominic-Madori Davis agreed to have her digital likeness made at Synthesia's new office in New York. Staff took numerous photographs in a small studio and recorded about 2 minutes of her voice. The company made both a personal likeness that reads supplied scripts and an interactive one that listens and answers, each with and without her glasses. Davis writes that Synthesia had previously made such a likeness only for corporate affairs lead Alexandru Voica. Her trial was the company's first with a journalist and its first beyond that employee, putting the product in front of a person whose work depends on the identity behind the voice.[1]
The interactive likeness returns to one investigation
The interactive version was tied to Davis's reporting on fraud at venture-backed companies. Speech is converted to text, a language model interprets the question, a reply becomes speech, and Synthesia's video model animates the face. The person on screen can therefore respond instead of merely reading a prepared script. Davis writes that questions beyond the selected investigation were directed back to that story. Her parents tried questions only the family would know and received no personal answers. The trial demonstrated a likeness with a deliberately narrow subject, not a copy that knows the reporter's life or can freely start new conversations under her identity.[1]
Synthesia places the likeness in three products
Davis describes three product routes at Synthesia: a creation platform that turns typed scripts into video, Sessions for interactive role practice, and an API for connecting the company’s audiovisual models with other services. Customers can choose speech and language components from Cartesia, ElevenLabs, Google, or OpenAI, and hosting can run in a customer's cloud or at Synthesia. Davis's likenesses were ready within days, but the finished voice lost the hoarseness of her recording; some friends thought it sounded unlike her. A convincing moving face and a recognizable voice were not identical outcomes in this particular trial, even though they emerged from the same capture session.[1]
Related columns
For more information on this topic, you can read the related columns.
