Only 2.2MB of Google’s speech recognition algorithm

US tech giant Google has many services. In some of these services, Vocal recognition features are used. For example, all the features like Assistant, Translation and Voice Recorder found on Pixel devices try to get the perception of speech in some way. However, achieving this was a very difficult problem. Aware of this situation, Google is now using a new file artificial intelligence model announced.

The biggest problem with applications that can recognize speech was extracting users’ voices from other speakers and noise in the environment. The company, which in 2018 intervened to overcome this problem, VoiceFilter He created an artificial intelligence called. Google, which has remodeled this artificial intelligence, which has a very heavy structure, is now only on devices. 2.2 MB appeared before consumers with the space-saving “VoiceFilter-lite”.

Artificial intelligence models for sound identification are generally heavy in nature. High battery consumption and lead to CPU usage. Here is Google’s new artificial intelligence model, which prevents it. With tons of mobile apps in harmony The artificial intelligence model that can work will provide various benefits to both Google and other developers.

According to statements made by Google VoiceFilter-liteuses the user’s already parsed voice. However, if a conflict occurs during detection, the AI ​​works to prevent this conflict. Furthermore, this artificial intelligence model does not need to be integrated into the main model. Of course, a previously unidentified voice can bypass VoiceFilter-lite and switch to a different voice recognition model. Just for the moment English Artificial intelligence, which can be used as a platform, is expected to support different languages ​​in the future.