Google is working to develop a new project, a single AI language model that supports the world's 1,000 most spoken or popular languages. This was revealed by Google’s senior VP Jeff Dean in a blog post where he discussed ‘3 ways AI is scaling helpful technologies worldwide’.
Universal Speech Model
In the blog Jeff said that as a first step towards this goal, the company has developed a Universal Speech Model (USM). USM is trained in more than 400 languages, which makes it the largest language coverage seen in a speech module. He also added that as Google expands its work, the tech giant is partnering with communities across the globe to source representative speech data.
With over 7,000 languages spoken in the world only a few are represented online today. This means that the traditional approaches used in training language models on text from the web fail to understand how we communicate globally and locally. Additionally, the traditional models of language also have flaws, such as re-enacting harmful social biases such as racism and xenophobia that the model fails to understand.
Read more: Pune's PMPML partnered with Google Cloud for real-time bus tracking
This is why, to remove the obstacle and ensure that the world's information is accessible and useful, Google has announced the 1,000 languages initiative. Google hopes that with this they can make the digital world more inclusive to the billions of people in marginalized communities.
Since the project is ambitious it will require inclusion across the world which could take many years to complete. However, Google is already on its way. The company recently launched voice typing for 9 more African languages by working hand-in-hand with researchers and organisations in Africa. They have also started working with local governments, NGOs and academic institutions in South Asia to eventually collect representative audio samples in regional languages and dialects.
Google is not the only company working on innovating new AI language models. Nvidia is also developing a new speech AI ecosystem through a partnership with Mozilla Common Voice. The company said that standard voice assistants, like Amazon Alexa and Google Home, support less than 1 per cent of the world's spoken languages and they plan to solve this problem by improving linguistic inclusion in speech AI and include speech data for other global and low-resourced languages.
