The latest addition to Google Search for mobile can understand your humming, whistling and singing to identify songs. Hum to search is triggered when you ask Google “what is this song?” and then proceed to perform bits of the tune to your best abilities. After you’re done, Google will present you actual songs based on a percentage match score which it thinks comes closest to your performance.
Google says its machine learning algorithms transforms the audio input into a number-based sequence which it then matches to other song melodies and tries to find the closest one. The model was trained with several sources with actual people singing, humming and whistling in addition to real studio recordings. Google’s algorithms can also strip away musical instruments, as well as the timbre and tone of the singer’s voice in order so it only relies on the actual numeric sequence in the song matching process.
Hum to search is now live on both Android and iOS Google Search apps and supports more than 20 languages.