Facebook/Meta developed a Hokkien AI translation system , which demonstrated a real-time translation between Hokkien and English. “We use Mandarin as an intermediate language to build pseudo-labels, we first translate English (or Hokkien) speech into Mandarin text, then we translate to Hokkien (or English) and add it to the training data.” Meta’s research Team said. Currently, the system allows Hokkien speakers to converse with English speakers, albeit bluntly. The model can only translate one complete sentence at a time. Meta announced that, in addition to the open-source models and training data obtained from this project, the company is releasing the first speech translation benchmark system based on the Hokkien corpus, as well as SpeechMatrix, “a database using Meta’s LASER data mining technology. Large Corpus of Speech-to-Speech Translation”, a system that will enable researchers to create their own speech-to-speech translation (S2ST) systems.
This article is reprinted from: https://www.solidot.org/story?sid=73132
This site is for inclusion only, and the copyright belongs to the original author.