Voice commands are gaining popularity because they are an easy and intuitive way to get answers, and ask smart devices to perform a task.
In 2018, the team unveiled the first system VoiceFilter Who uses the company’s voice-matching technology.
It is used in the Google Voice Assistant to analyze your speech and voice.
Google has a lot of applications within your phone that use voice recognition technology, such as: Google Assistant, Google Translate, and voice recording app.
These apps face the challenges of separating your voice from other people or separating out background noise.
VoiceFilter has been touted to excel in achieving a better source-to-distortion ratio (SDR), which helps it recognize sounds more efficiently than traditional methods.
More efficient voice recognition and better source-to-distortion ratio SDR requires a large system, while consuming battery and CPU power.
Google believed that the system would be faster and better if it worked within the device, even in offline mode.
The VoiceFilter Lite – only 2.2MB in size – is intended for in-device use to significantly improve speech recognition in spoken speech.
The system is compatible with a large number of mobile device applications by recognizing the recorded voice of a specific speaker.
The new system must be able to recognize the user’s voice in very noisy conditions, as well as when an Internet connection is not available.
According to Google, the system was able to provide a 25.1 percent improvement in the word error rate (WER) – a ratio of how many words the system can recognize from a reference statement – for overlapping speech.
And apps can get much better at capturing voice commands with the new VoiceFilter-Lite model.
It should also help the system process queries faster, given that it works offline.
VoiceFilter-Lite offers promising results, but it currently only works with English, and Google is working on adopting the system with other languages as well.