News
News Categories

Meta’s AI speech translation system hits its first milestone – by translating Hokkien

By Liu Hongzuo - on 20 Oct 2022, 12:18am

Meta’s AI speech translation system hits its first milestone – by translating Hokkien

Meta's AI Universal Speech Translator finally has its first milestone -- translating Hokkien.

Meta is on a roll with AI-related advancements, having announced its Grand Teton platform earlier and it’s now flexing the very first milestone of Meta’s AI Universal Speech Translator (UST). Meta's AI UST can now translate Hokkien.

Hokkien as a spoken dialect in Singapore needs very little introduction to Singaporeans. If you lived long enough here, you’d develop an intimate understanding of certain colourful words from the language.

Source: Meta.

According to Meta’s latest blog post, Hokkien as an oral language has no standardised written form despite being widely spoken within the Chinese diaspora – such as Taiwan, Singapore, and certain parts of China. The lack of a standard writing form makes it harder for Hokkien speakers to pass down the language intact, and also creates a communication barrier between different generations of speakers.

Typically, translation requires writing, but the development of Meta’s AI UST’s first translation milestone makes it possible for Meta to translate spoken English into Hokkien, and vice versa. As demonstrated in the clip above (starring Mark Zuckerberg), the translation does not require transcoding into writing. Most listeners with a rudimentary grasp on Hokkien would also be able to tell that the translation is miles cleaner than what you'd usually hear in a coffee shop.

The Meta AI UST project is an open-source, speech-to-speech translation system that eventually hopes to achieve real-time, speech-to-speech translation “across all extant languages”. There are approximately more than 7,000 living languages that are primarily oral like Hokkien, making this just the very first step Meta is taking for real-time speech translation.

Source: Meta.

The technical explanation for AI-based Hokkien translation can be found here. Some interesting challenges cropped up in Meta’s pursuit towards making Hokkien translation possible, and it includes having to translate into Mandarin first, having to encode Hokkien speech embeddings to pretrain AI, and having to do all this without written translation as a resource.

Join HWZ's Telegram channel here and catch all the latest tech news!
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.