As Microsoft and Google butt heads over whose AI chatbot is better, that isn’t the one use for machine studying and language fashions. Together with rumored plans to indicate off greater than 20 merchandise powered by synthetic intelligence throughout its annual I/O occasion this yr, Google’s progressing towards its purpose of constructing an AI language mannequin that helps 1,000 completely different languages. In an update posted on Monday, Google shared extra details about the Common Speech Mannequin (USM), a system Google describes as a “important first step” in realizing its targets.
Final November, the corporate announced its plans to create a language model supporting 1,000 of the world’s most-spoken languages whereas additionally revealing its USM mannequin. Google describes USM as “a household of state-of-the-art speech fashions” with 2 billion parameters educated on 12 million hours of speech and 28 billion sentences throughout over 300 languages.
USM, which YouTube already makes use of to generate closed captions, additionally helps automated speech recognition (ASR). This routinely detects and interprets languages, together with English, Mandarin, Amharic, Cebuano, Assamese, and extra.