Google shared additional information about the Universal Speech Model (USM), which the company described as a “critical initial step” toward realizing its goals. It is now closer to its goal of creating an AI language model that can support 1,000 languages to defeat ChatGPT.
In November, the company revealed its USM model and announced plans to create a language modeling system that supports 1,000 languages.
The tech giant describes USM as a collection of state–of–the–art speech models that were trained on 12M hours of speech and 28B sentences of text, spanning 300+ languages.
Google stated in a blog post that USM, which can be used on YouTube (e.g., for closed captions), can perform automatic Speech Recognition (ASR) on both widely spoken languages like English, Mandarin, and Cebuano as well as on languages with low resources like Amharic and Cebuano.
Google claims that USM currently supports more than 100 languages and will be the foundation for a larger system.
Google will soon introduce an array of AI features to its products. Gboard for Android, for example, is currently working to integrate Imagen’s text-to-image generator.