Let your voice out! Google will show what the song is only with your voice humming – 10/16/2020


Who has never had a song in their head and spent days and days humming its melody? Because Google announced on Thursday (15), during the global event “Search On”, a new resource to help in this search.

If the microphone feature of your mobile app – or Google Assistant – was able to recognize a song that was playing in the environment before, now you don’t need that much information. The promise is that the new “hum to search” tool will allow you to simply hum (there-only-there), whistle or sing a little bit of the song to solve this mystery.

The tool is only available in English on iOS and in more than 20 languages ​​on Android, with expansion plans. According to the instructions, you must open the latest version of the Google application, press the microphone icon and say “what’s this song?” (what song is that ?, in free translation), or click on the “Search a song” button. Then, just start singing for ten to 15 seconds.

Image: Reproduction / Google

Google Assistant should do the same: say “Hey Google, what’s this song?” (Hey, Google. What song is that?) And then show what you know about the song. The search result will show songs that can match the melody you performed.

Tilt did the step by step using the application on a smartphone with Android operating system and settings in Portuguese. We asked “what song is that?” which actually triggered the feature. We tried to hum “Me Gusta”, the new hit of singer Anitta (with lyrics in English and Spanish) but the app did not recognize the song, although it presented another option with percentages of correspondence. We repeated the test with “Evidências” by Chitãozinho and Xororó and the app identified it at first, albeit with other interpreters.

According to Google, the feature works by using its machine learning models to “turn the audio into a number-based sequence that represents the melody of the song”. In simpler language, it is like a “fingerprint” of the music, which can be compared to the songs in your database.

The company explains that it trains these models in “a variety of sources, be it people singing, whistling or humming, as well as studio recordings”, removing instruments and vocal quality to focus only on this numerical sequence. That is: in theory, tuning is not a requirement for you to be successful in searches.


Please enter your comment!
Please enter your name here