Ever had a tune stuck in your head but couldn’t remember the song’s name or lyrics? Now, you can Find That Song simply by humming, whistling, or singing it to Google Search. This innovative feature leverages the power of machine learning to identify melodies, acting like a musical fingerprint scanner for songs.
The Melody Fingerprint: How Machines Learn to Recognize Music
Just like every person has a unique fingerprint, every song possesses a distinct melodic identity. To find that song from your hum, our machine learning models are designed to decipher this “melody fingerprint”. When you hum into Google Search, the audio is transformed into a number-based sequence. This sequence is essentially a digital representation of the song’s melody.
Our models undergo extensive training, learning to recognize songs from diverse audio inputs. Whether it’s someone singing, whistling, humming, or even a studio recording, the system can adapt. Crucially, the algorithms are engineered to filter out extraneous elements like instruments, vocal timbre, and tone. By stripping away these layers, what remains is the core melody, the song’s unique number-based fingerprint, which is essential to find that song.
Real-Time Song Matching: Identifying Your Hummed Tune
Once your hum is converted into a melodic fingerprint, the real magic happens. We compare this sequence against a vast database of thousands of songs from across the globe. This comparison occurs in real-time, rapidly identifying potential matches to find that song you’re searching for.
Think of a popular song like Tones and I’s “Dance Monkey.” You can instantly recognize it whether it’s sung, whistled, or played in its original studio version. Similarly, our machine learning models are trained to recognize the underlying melody of the studio recording. This allows the system to effectively match your hummed audio to the correct song, helping you find that song effortlessly.
Built Upon Years of Music Recognition Innovation
This humming-to-search feature isn’t built overnight. It’s a significant step forward, expanding upon our Research team’s earlier breakthroughs in music recognition technology. We first introduced low-power, on-device music recognition with Now Playing on Pixel 2 in 2017, utilizing deep neural networks. In 2018, we extended this technology to the SoundSearch feature within the Google app, broadening its reach to a catalog of millions of songs.
This new humming feature takes music identification to the next level. Now, to find that song, you don’t need lyrics or even the original song playing nearby. All it takes is a simple hum, and the power of machine learning helps you uncover the music you’re looking for.
