Ever had a tune stuck in your head but couldn’t quite place the song? We all know the frustration of trying to describe “what’s the song that goes” with a certain melody to friends or even search engines, often with little success. But what if you could simply hum, whistle, or sing that melody and instantly find the song you’re looking for? Thanks to advancements in machine learning, this is now a reality.
Think of a song’s melody as its unique fingerprint. Just like fingerprints differentiate individuals, melodies distinguish songs. Google has developed innovative machine learning models capable of recognizing these melodic fingerprints. When you hum a tune into the search feature, these models spring into action, transforming your audio input into a numerical sequence. This sequence acts as the melodic fingerprint of your hummed tune.
These sophisticated models are trained on a vast dataset of songs, learning to identify them from various sources. Whether you’re humming, whistling, singing, or even if it’s a studio recording, the underlying melody can be extracted and recognized. Crucially, the algorithms are designed to filter out extraneous details like accompanying instruments, vocal timbre, and tone. This process isolates the core melody, leaving behind a clean, number-based sequence – the song’s melodic fingerprint.
This melodic fingerprint is then compared in real-time against a massive database containing thousands of songs from across the globe. The system swiftly identifies potential matches, bringing you closer to discovering “what’s the song that goes” in your head. Consider the popular song “Dance Monkey” by Tones and I. You’d recognize it whether you heard the studio version, a live performance, or someone simply humming the tune. Similarly, the machine learning models recognize the core melody across these variations, enabling a match even from a simple hum.
This groundbreaking feature builds upon previous work in music recognition technology pioneered by Google Research. The journey began with “Now Playing” on Pixel 2 in 2017, which brought low-power, on-device music recognition using deep neural networks to mobile. In 2018, this technology was further integrated into the SoundSearch feature within the Google app, expanding its reach to a catalog of millions of songs. This new hum-to-search experience represents a significant leap forward, empowering you to identify songs based purely on melody, even without lyrics or any original recording. Now, all it takes is a hum to finally answer the question: “what’s the song that goes like that?”
