Frequently, people find themselves unable to recall the lyrics or even the title of a song, even though the melody is firmly imprinted in their mind. The frustration of such moments is universal. However, today, at Search On, we are thrilled to announce that Google can now help you resolve this issue – no need for lyrics, knowledge of the artist, or perfect pitch.

Starting today, all you need to do is hum, whistle, or sing a part of the melody to Google, and it will help you identify that elusive tune. On your mobile device, simply open the latest version of the Google app or locate the Google Search widget, tap the microphone icon, and say, “What's this song?” or click the “Search a song” button. Then, hum the melody for about 10-15 seconds. If you're using Google Assistant, it's just as straightforward. Say, "Hey Google, what’s this song?" and proceed to hum the tune. Currently available in English on iOS and in more than 20 languages on Android, we aim to expand this feature to even more languages in the future.

Once you've finished humming, our machine learning algorithm goes to work, helping identify potential song matches. You don't need to worry about having perfect pitch – we'll present you with the most likely options based on the melody. You can then select the best match and explore comprehensive information about the song and artist, watch any accompanying music videos, listen to the song on your preferred music app, find the lyrics, read analyses, and even check out other recordings of the song if available.

But how does it all work? To put it simply, a song's melody is like its unique fingerprint. We've developed machine learning models that can match your hum, whistle, or singing to the correct "fingerprint." When you hum a melody into Search, our models transform the audio into a number-based sequence representing the song's melody. These models are trained to recognize songs based on a variety of sources, including human singing, whistling, humming, and studio recordings. They filter out all other details like accompanying instruments, timbre, and tone of the voice, leaving only the number-based sequence or "fingerprint" of the song.

We then compare these sequences to a vast database of songs from around the world and identify potential matches in real time. For example, if you're humming a bit of Tones and I's "Dance Monkey," our models will recognize the song whether it was sung, whistled, or hummed. They can even recognize the melody from the studio-recorded version of the song and match it with your hummed audio.

This innovative feature builds upon the foundation of our Research team's music recognition technology. We initially introduced "Now Playing" on the Pixel 2 in 2017, employing deep neural networks to enable low-power music recognition on mobile devices. In 2018, we expanded this technology to the SoundSearch feature in the Google app, extending our database to include millions of songs. This new experience represents a significant leap forward because it allows us to recognize songs even without lyrics or the original track – a mere hum is all that's required.

So the next time you find yourself unable to remember the name of a catchy song you heard on the radio or that beloved classic your parents adore, just start humming. Your answer will be delivered in record time.