Google, the tech giant, has provided an update on its Universal Speech Model (USM), which was announced in November 2021. As the competition in AI heats up, Google’s ambitious plan is to support the world’s most commonly spoken languages through its state-of-the-art family of speech models with 2 billion parameters. The models were trained on 12 million hours of speech and 28 billion sentences of text across over 300 languages.
Although Google has not achieved its aim of supporting 1,000 languages yet, the USM is a significant milestone for the company as it competes against ChatGPT, which is backed by Microsoft. Google currently uses the USM to generate live captions for YouTube videos.
The company claims that the model can identify not only widely spoken languages such as English and Mandarin but also under-resourced languages like Amharic, Cebuano, Assamese, and Azerbaijani, among others.
The building blocks of the USM will serve as a foundation for Google to build an even bigger language model in the future.
“The development of USM is a critical effort towards realizing Google’s mission to organize the world’s information and make it universally accessible,” wrote Google in the blog post.
