Ever had a tune stuck in your head, but you just couldn’t place the song? You hum it, maybe whistle a few bars, hoping someone recognizes it. Now, imagine technology that can understand your hum and tell you “what’s this song hum” is. That’s the power of machine learning at work, turning your vocal melodies into instant song identification.
At its core, a song’s melody is its unique identifier, much like a fingerprint. Each melody possesses a distinct pattern. We’ve developed sophisticated machine learning models trained to recognize these melodic fingerprints. When you hum, whistle, or sing into a search engine, these models spring into action, transforming the audio into a numerical sequence. This sequence acts as the song’s “fingerprint.”
These aren’t just any models; they’re trained on a vast dataset of songs and various vocal inputs – from studio recordings to people humming and whistling. Crucially, the algorithms are designed to filter out extraneous details like instrumental accompaniment, vocal timbre, and tone. This isolation process leaves only the essential melodic fingerprint, the number-based sequence that defines the song.
The magic happens when this melodic fingerprint is compared against a massive database of songs from across the globe. The system identifies potential matches in real-time, sifting through thousands of songs to find the best fit. Think about a popular song like Tones and I’s “Dance Monkey.” You instantly recognize it whether you hear the studio version, a live performance, or even someone humming it. Similarly, our machine learning models can recognize the underlying melody of the studio recording and match it to your hummed input.
This technology is a direct evolution of our research team’s prior work in music recognition. Starting with “Now Playing” on Pixel 2 in 2017, we utilized deep neural networks for low-power, on-device music identification. This capability expanded to the Google app’s SoundSearch in 2018, reaching a catalog of millions of songs. However, this new hum-to-search feature marks a significant leap forward. Now, you don’t need lyrics or even the original song playing nearby. All it takes is your hum to uncover “what’s this song hum” you’ve been wondering about.
