Do you remember the album “da daaaa da da daaaa na naa naa ooohh yeah”? Or the one that starts with the guitar chords going, “da na na naa”? We've also experienced the frustration of forgetting the name of an album or any of its lyrics but seeing the melody trapped in our heads. We revealed today at Search On that Google will now help you find it out without the need for songs, artist names, or perfect pitch.

Hum to search for your earworm

To solve the earworm, you can now buzz, whistle, or sing a song to Google. Open the new update of the Google app or find the Google Search widget on your mobile screen, press the microphone key, and say "what's this song?" or press the "Search an album" button. Then, for 10-15 seconds, begin humming. It's the same way on Google Assistant. Say something like, "Hey Google, what's this song?" and then hum it. This functionality is currently available on iOS in English and Android in over 20 languages. In the future, we plan to extend this to other languages.

Our machine learning algorithm helps recognize possible music matches once you've stopped humming. Don't worry, perfect pitch isn't needed to use this feature. Based on the song, we'll show you the most possible choices. Then you can choose the best fit to learn more about the album and artist, watch any accompanying music videos, listen to the song on your favorite music app, find the lyrics, read commentary, and, if appropriate, check out other recordings of the song.

If you have a song trapped in your head? To hunt, simply hum


How machines learn melodies 

So, how does it function? A simple analogy is that a song's melody is similar to its fingerprint: each one has its own distinct identity. We've created machine learning models that can detect the correct "fingerprint" in your hum, whistle, or singing.

When you hum a melody into Search, our machine learning models convert the sound into a number-based sequence that represents the melody of the album. Our algorithms have been equipped to recognize songs from a variety of sources, including human singing, whistling, and humming, as well as studio recordings. All other information, such as accompanying instruments and the timbre and tone of the voice, are also removed by the algorithms. The song's number-based list, or fingerprint, is all that's left.

We equate these sequences to thousands of songs from all over the world in real-time, looking for possible matches. If you listen to Tones and I's "Dance Monkey," for example, you'll hear it if it's chanted, whistled, or hummed. Our machine learning models also know the rhythm of the studio-recorded version of the album, which we can use to adapt it to hummed music.

This builds on the music recognition technologies developed by our Research team. In 2017, we released Now Playing on the Pixel 2 to introduce low-power music recognition to mobile devices using deep neural networks. In 2018, we applied the same technologies to the Google app's SoundSearch functionality, allowing it to access a database of millions of tracks. This recent experience takes things a bit further, as we can now remember music without listening to the words or listening to the original album. We just require a hum.

If you have a song trapped in your head? To hunt, simply hum


So, the next time you can't recall the name of a catchy song you heard on the radio or a retro tune your parents like, start humming. You'll get your response in a flash.



Post a Comment

Previous Post Next Post