Of the approximately 7,000 known languages in the world, 40% do not have a standard writing system. These textless languages pose a unique problem for modern machine-learning translation systems, since translation often requires converting spoken language into written text, and then translating the text back to speech.
However, Meta announced on October 19 that it has solved this problem with its latest open source language artificial intelligence (AI).
Meta has unveiled “the first artificial intelligence-driven speech translation system for a textless language,” which successfully translated Hokkien, a spoken language primarily spoken in Taiwan.
The system is part of Meta’s artificial intelligence project, dubbed the “Universal Speech Translator (UST).”
“The translation system is the first milestone in Meta AI’s Universal Speech Translator project, which focuses on developing artificial intelligence systems that provide real-time speech-to-speech translation in all languages, even without Meta,” Meta said in a statement. The language of words.” The project is working to develop more real-time speech-to-speech translations so that the inhabitants of the metaverse can interact more easily.
Most previous AI-driven language translation efforts have focused on written languages, but one of its goals is to translate more languages that lack formal writing systems, Meta said.
According to Meta, there are more than 7,000 languages, but “more than 40 percent are primarily spoken, with no standard or well-known writing system.”
Meta CEO Mark Zuckerberg demonstrated the use of the system to translate Hokkien in a video posted on Facebook, and said Meta will open source the tool so that people can use the new artificial intelligence system to translate more languages.
