The development of artificial intelligence automatic translation systems is all, and the effect has been getting better and better, but the current technology is still based on text for translation, not directly through voice. Recently, Meta has developed a textless translation system. Even if there is no standard text writing system, AI models can still be trained.
Meta said that the newly launched “Universal Speech Translator” (UST) is the world’s first artificial intelligence “spoken-to-speak” translation system, which can directly correspond to languages without standard writing systems and lack of professional translators. translate.
(1/3) Until now, AI translation has focused mainly on written languages. Universal Speech Translator (UST) is the 1st AI-powered speech-to-speech translation system for a primarily oral language, translating Hokkien, one of many primarily spoken languages. https://t.co/onYKQ8uoKN pic.twitter.com/Iy8MRMOypQ
— Meta AI (@MetaAI) October 19, 2022
The first language they used this technology chose a combination of Taiwanese (Hokkien) and English. The system can already automatically translate and pronounce the dialogue between the two languages.Mark Zuckerberg demonstrated the technology himself. Meta said that for this oral language, it is quite difficult to collect data, design models and evaluate accuracy. It is necessary to first analyze the voice using “voice mining” technology, then pair Hokkien and Chinese, and then translate it to English to increase. Go to the training database, and add your own pinyin system in the middle to evaluate the accuracy. Meta said the technology could be extended to other languages in the future, and the “Large Speech Translation Corpus” (LASER) will be made public for other researchers to use.