Do you remember the song “da daaaa da da daaaa na naa naa ooohh yeah” that goes? Or the one that begins with the “da na na naa” guitar chords? When you do not remember the name of a song or any of the words, but the melody is stuck in your brain, we all know how frustrating it is. We announced today at Search On that Google will now help you find it out, with no lyrics, artist name or perfect pitch required.
You should hum, whistle or sing a melody to Google to solve your earworm, starting today. “Open the new version of the Google app on your mobile device or find your Google Search widget, tap the mic icon and say,” What’s this song? “or click on the button” Check for a song. Start humming for 10-15 seconds after that. It is just as fast on Google Assistant. Say, “Hey Google, what song is this?” “And then the tune would hum. This feature is currently available on iOS in English and on Android in over 20 languages. And in the future , we plan to extend this to more languages.
Our machine learning algorithm helps classify possible song matches after you’re done humming. And don’t worry, to use this feature, you don’t need the perfect pitch. Based on the tune, we’ll show you the most likely choices. You can then choose the best match and explore song and artist content, view any accompanying music videos or listen to the song on your favourite music app, find the lyrics, read review, and even check out other song recordings when they are available.
And how is it working? An simple way to describe it is that the melody of a song is like its fingerprint: they each have a distinctive identity of their own. We’ve developed machine learning models that can fit the right “fingerprint” with your hum, whistle or singing.
Our machine learning models turn the audio into a number-based sequence representing the melody of the song as you hum a melody into Quest. Our models are equipped to recognise songs based on a number of references, like singing, whistling or humming by humans, as well as recordings from the studio. All the other data, such as accompanying instruments and the timbre and tone of the voice, are also taken away by the algorithms. The number-based sequence of the song is what we’re left with, or the fingerprint.
We equate these sequences with thousands of songs from around the globe and in real time recognise possible matches. If you listen to Tones, for instance, and I’m ‘Dance Monkey,’ you can remember the song whether it’s sung, whistled, or hummed. Similarly, the melody of the studio-recorded version of the song that we can use to align it with a person’s hummed audio is recognised by our machine learning models.
This builds on the work of the music recognition technologies of our AI Research team. In 2017, we introduced Now Playing on the Pixel 2, using deep neural networks to carry low-power music recognition to mobile devices. In 2018, in the Google app, we took the same technology to the SoundSearch feature and extended the reach of millions of songs to a catalogue. The new experience takes things a step further, because without the lyrics or original tracks, we can now recognise tracks. All that we need is a buzz.
So next time you can’t remember the name of a catchy song you’ve heard on the radio or your parents love that classic jam, just start humming. In record time, you’ll have your answer.