{"id":111053,"date":"2020-11-12T15:16:09","date_gmt":"2020-11-12T12:16:09","guid":{"rendered":"https:\/\/en.buradabiliyorum.com\/at-just-2-2mb-googles-new-speech-filtering-tech-is-perfect-for-mobile-apps\/"},"modified":"2020-11-12T15:16:09","modified_gmt":"2020-11-12T12:16:09","slug":"at-just-2-2mb-googles-new-speech-filtering-tech-is-perfect-for-mobile-apps","status":"publish","type":"post","link":"https:\/\/buradabiliyorum.com\/en\/at-just-2-2mb-googles-new-speech-filtering-tech-is-perfect-for-mobile-apps\/","title":{"rendered":"#At just 2.2MB, Google\u2019s new speech filtering tech is perfect for mobile apps"},"content":{"rendered":"

“#At just 2.2MB, Google\u2019s new speech filtering tech is perfect for mobile app<\/a>s<\/strong>”
\n<\/p>\n

\n Google has plenty of apps on your phone that use speech detection from Google Assistant to Google Translate and Pixel\u2019s nifty recorder app. However, one of the challenges these apps face is separating your voice from other people or background noise.<\/p>\n

To overcome these challenges, Google\u2019s AI team has built a new lightweight model called VoiceFilter-lite. In 2018, the team unveiled the first VoiceFilter<\/a> model which used the company\u2019s voice match tech. It\u2019s used in Google Assistant to analyze your speech and sound when you enroll for a\u00a0service.<\/p>\n

A lot of times recognizing voices efficiently \u2014 technically, achieving better source-to-distortion ratio (SDR) \u2014 takes a model with a large size, prominent CPU power, and battery consumption.<\/p>\n

That\u2019s why the Google team came up with the VoiceFilter-lite model that\u2019s sized at just 2.2MB, making it suitable for tons of lightweight mobile applications.<\/p>\n

It uses the already enrolled voice of a user and improves the recognition even when there\u2019s overlapped speech. Google claims the model enhances recognition by 25% word error rate (WER)<\/a> \u2014 a ratio used to measure how many words a model recognizes from reference sentence
\n