November 8, reporters learned that Ali Baba Dharma School of Machine Intelligence Laboratory latest research --KAN-TTS will be the first large-scale applied this year Lynx double 11, based on this technology, rookie hotline robots, robot voice Xiaomi and Lynx close the wizard will provide live voice interactive experience consumers around the world.
Let machines speak is one of the basic techniques of artificial intelligence, dating back to the birth of TTS (Text To Speech) technology in 1960, but let the machine issued a vivid sound have always been a problem, it is understood that the traditional voice synthesis requires massive text and audio information, the synthesized speech and the original audio closeness only between 85% to 90%.
In July, Dharma hospital released a new generation of speech synthesis technology KAN-TTS, the first increase this number to more than 97%. This is believed to be named MIT Technology Review 2019 years after the "Global Top Ten breakthrough technology," Ali Baba speech technology to enhance the strength of the leap again.
Transfer learning algorithms and a variety of new model, KAN-TTS can quickly generate specific speakers according to the style of highly similar voice, speech synthesis and greatly reduced the threshold, recording phone ten minutes, the machine can be completed by the algorithm to mimic the sound.
Over the past few months, KAN-TTS technology has achieved full coverage of the mainstream scene style sound can be for common scenarios, customer service scenarios, childish scene, the English scene and dialects scene, available in 41 high-quality sound, such as gentle, sweet, harsh and other styles. According to Bodhidharma hospital experts said the team also plans to use the technology to help the visually impaired and speech impaired barrier-free communication.
Since the establishment of Dharma hospital for two years, Ali Baba in vision, speech and natural language processing and other fields has set a number of world records, and emerged as China's largest artificial intelligence company. This year's conference Yunqi Hangzhou, Alibaba said Ali AI calls over 1 trillion a day, serving 1 billion people worldwide, the daily processing image 1 billion, 1.2 million hours of video, voice, natural language 550,000 hours and 500 billion sentence.