News anchors(主播) must have been reluctant to read out the following news: Xin Xiaomeng began working as the world's first female artificial(人工的) intelligence news anchor at Xinhua News Agency on Sunday, three months after a male robot joined the profession.
Unlike previous news robots though, Xin does not read news like a cold machine; she reads it almost like a human being. The muscles on her face stretch and relax-and her reactions change-as she continues reading. That's why many news anchors were worried: Will AI replace us in the near future?
To find the answer, we have to analyse the technologies that support Xin at her job. Three key technologies are used to support Xin. First, samples of human voices are collected and synthesized (合成). This is followed by the collection and synthesis of human muscle movement samples. And third the voices and movements are married in a way that when the Al news anchor reads, the micro -electric motors behind her face move to make her expressions seem more human.
Yet we need a thorough knowledge of deep leaning technology to make a robot imitate a person's voice. The developer needs to collect tens of thousands of pieces of pronunciations, input them into the machine and match them with the text or the Al to lean and read. The process for imitating facial movements is similar. The developer has to analyse the movements of the 53 muscles in the human face, make a model set from the collected data for the AI news anchor to lean, and imitate the movements of facial muscles via programs
Both the technologies used to make Xin's performance impressive are mature. The real difficulty lies in the third -the technology to match the pronunciations with facial movements so that Xin expressions vary according to the content of the news report. In fact, Xins expressions don't always change according to the content. As a result, her expressions look anything but human. Actually. AI is still no match for human qualities.
