Google has rolled out an update to its new Universal Speech Model (USM) to battle with Open AI. In reality, it is a speech model which supports 1,000 languages.
Related to the Universal Speech Model, the researcher reveals that it surpassed Open AI for parts of automated speech recognition. For instance, it can write the best youtube captions.
What Is The Universal Speech Model Of Google?
The Universal Speech model contains various speech models having two billion parameters. And, it has been trained where a big dataset of text including 12 million hours of speech as well as 28 billion sentences targeting more than 300 languages.
credit:G00gle
The particular model used in YouTube can do speech recognition on its own not only on just broadly spoken and understood languages but also in some languages which are not spoken widely such as Amharic, Cebauano, Assamese, and Azerbaijani.
As per Google, the USM can identify over 1000 languages grounded on automatic speech recognition scaling. Meanwhile, multilingual datasets can also be recognized by this model.
As per the company, by this model, the encoder can pre-train and very well adapt to new languages and data in the course of training. With the help of the model, data from various languages can be swiftly filtered and can also be utilized for content as well as speech.
credit:G00gle
The researchers also showed with the model the effectiveness of the pre-trend encoder by fine-tuning on multilingual motion data of YouTube captions.
In spite of the restricted data of YouTube, the model made not more than 30% of word faults on average in 73 languages, a feat never attained before it. As per the researchers, USM Automation surpasses OpenAI in every aspect of speech identification.
The Machine Learning of Google
Google has utilized about 1,000 languages that are widely spoken in the world to make a machine-learning model. It was first rolled out in November 2022. Although, some of these languages are spoken by not more than 20 million people.
credit:google
Basically, Google is accumulating data on the basis of their restricted available data and few speakers. In a machine learning model, universal speech is made by accumulating data from about all languages.
This clearly means that the new Universal Speech Model of Google will be capable of giving more precise detail in all languages.
credit: G00gle
ALSO READ: Snapchat Has Launched Its Own Chatbot, Know How To Access
It is very interesting to know this update from Google. It will open the doors of more flexibility related to the languages spoken over the world.
ALSO READ:US Universities Changing Their Studies Patterns Due to AI Chatbots Use