Have you ever had a tune stuck in your head but couldn’t remember the song’s name? Now, thanks to advancements in machine learning, you can simply hum, whistle, or sing that melody to Search and discover the song you’re looking for. This innovative feature uses the unique melodic “fingerprint” of a song to identify it, even without lyrics or instrumentation.
At the heart of this technology are sophisticated machine learning models designed to recognize melodies. Think of a song’s melody as its unique identifier, much like a fingerprint. When you Hum A Song into Google Search, these models spring into action, transforming your audio input into a sequence of numbers. This numerical sequence represents the core melody of your hummed tune. Crucially, the system is trained to recognize songs from various sources – whether it’s someone humming a song, whistling a catchy tune, singing their favorite part, or even from studio recordings.
The algorithms are also designed to filter out extraneous details. Elements like accompanying instruments, the timbre of your voice, and your vocal tone are disregarded. The focus narrows down to the essential melodic contour of the song. What remains is that crucial number-based sequence – the melodic fingerprint – ready for comparison.
This is where the magic happens. Your hummed melody’s “fingerprint” is then compared against a vast database of thousands of song fingerprints from around the globe. The system works in real-time to identify potential matches. Consider the popular song “Dance Monkey” by Tones and I. You can recognize it instantly whether you hear the studio version, someone singing it, or even just a whistle of the tune. Similarly, the machine learning models recognize the underlying melody, allowing them to connect your hummed version to the official studio recording and identify the song.
This capability builds upon previous breakthroughs in Google’s music recognition technology. It’s an evolution of the technology behind features like Now Playing on Pixel devices, launched in 2017, and SoundSearch in the Google app in 2018. Those earlier iterations utilized deep neural networks for low-power music recognition, expanding to a catalog of millions of songs. However, this new “hum to search” experience represents a significant leap forward. Now, song recognition is possible without relying on lyrics or original recordings – all it takes is your hummed melody to unlock the song you’re searching for.
