Of the Google Pixel 4, many things came to attention, but among all its novelties, it stood out, above all, a new application introduced by Google, which took advantage of the advances in terms of machine learning achieved by Google to evolve a task so apparently simple as it is to record our voice or that of other people.
And is that the recorder app of the Pixel 4, now available in the other models of the Pixel family, surprises with its ability to generate transcriptions in real-time without the need for an Internet connection, and even allow the search of text between all Recordings stored in an extremely precise way. And now, this exclusive Google Pixel app can be on your mobile, whatever the brand.
A modified version of the Google recorder that works on any mobile
The task of modifying the Google application to make it work on most Android terminals has been the work of the people of XDA-Developers. In theory, the modified version should work on devices of brands such as Huawei and Honor, LG, Motorola, Nokia, Samsung or Sony, in addition to those terminals with modified AOSP ROMs based on the Android 9 or Android 10 version. On the other hand, The app seems to have problems in ASUS terminals with Android 9 or 10, as well as OPPO, Realme and OnePlus.
As in the original application, the only language supported so far for transcriptions is United States English. Also, in this modified edition, the audio quality seems not to be as good as in the version available for Google models.
The app in question can be downloaded from the XDA thread completely free. However, it is also possible to resort to the original version that was leaked a few days before the presentation of the Pixel 4, and that still works on some devices beyond the latest models of the great G brand.
What’s so special about the Pixel recorder?
Now, what is really special about this app compared to the thousands of recorders that exist in Google Play? Google itself shared a few days ago a publication in its AI blog, explaining how the machine learning systems integrated in the device that give life to this application and its various functions work.
According to Google, to develop the real-time transcription function, an offline voice recognition system model is used, capable of transcribing recordings long enough – up to several hours long – in the most reliable way possible, while it is indexed, assigning words to timestamps calculated by the voice recognition model.
Thus, when we touch a word in the transcript, playback starts from the point we have selected. This system also allows searches between transcripts to go to an exact point.
Beyond that, Google explains that the app uses a neural network system capable of associating different sounds, such as barking, music or the sound of a specific musical instrument, with colors, to offer the user the possibility of visually recognizing each sound and that when exploring a recording they can skip those parts considered less important.
Finally, the application is also capable of recommending titles or labels based on the content of the recording, all without sending any data to Google’s servers. For this, an algorithm is used that collects nouns and proper nouns that are usually remembered more easily by users, and are organized according to their relevance, to finally offer the user a selection of words that can be used as titles or labels to identify more Easily every recording.