Appearing almost from the beginning of the Apple App Store, music identification applications are truly amazing... how can they recognize a song in just a few seconds using your phone's microphone?
Under the hood of its apps are sophisticated algorithms that allow you to find songs in an instant. Here's how it works!
The Magic of Music Identification Apps
If you use this type of app, you will certainly recognize yourself in this situation. You are driving in your car or having dinner in a nice restaurant, when you suddenly hear a song that you like and that you would like to listen to again later. So, you take out your phone, open Shazam and hold your device towards the sound source. In the blink of an eye, the application finds the name of the song, its artist and on which platform you can find it.
This type of app is fast, remarkably accurate, and can identify even the most obscure songs. In short, the app works by isolating the song and searching for it in a vast database of songs. But the technology these apps rely on to do this is quite complex and impressive.
The most amazing thing is that Shazam was released in 2002, and the system was as accurate and fast then as it is today. All thanks to a unique algorithm that would revolutionize the world of music.
At first glance, music identification apps like Shazam may seem simple. You might think they simply listen to lyrics, like any voice assistant, and search through a database of song lyrics to tell you what song you're listening to.
But it's not that simple. Most music identification apps are able to identify instrumental songs and find the singer of covers. Indeed, instead of analyzing the Lyrics of the song, they identify "fingerprints" specific to each song before comparing them with what they have in their huge database.
Identify the fingerprint of a song
You probably have devices that can be unlocked using your fingerprint. This is made up of little lines that are unique to you. Shazam works on a similar principle. The pieces are transformed into a data pattern that Shazam or another app can look up in its database.
At first glance, this method seems prone to several problems. Most of the time, when you hear music in public, there is background noise and distortion caused by the speakers, which can make songs unidentifiable or lead to inaccurate matches. Additionally, there is a lot of data captured in even a short piece of music, which can slow down the search for these patterns in a database of millions of songs.
In an interview with Scientific American In 2003, Avery Li-Chun Wang, Shazam's chief data officer and co-founder, explained how their algorithm solved these problems. The information in an audio sample can be visualized using a 3D graph called a spectrogram, which represents changes in frequencies over a period of time and can also display the loudness (amplitude).
In the same way that humans can only perceive sounds within a particular frequency spectrum, instead of considering the entirety of a song when searching, Shazam only considers frequency “peaks.” So the fingerprints it captures only consider the highest frequency points in a given time frame, and then the peak amplitude points within those frequencies.
In a research paper for Columbia University, Wang says the method was able to remove most of the unnecessary parts of an audio clip, such as background noise, and remove distortion. It also made the fingerprint size small enough that it took just a few milliseconds to identify a song from their vast database.
Since the advent of the precursor Shazam, a handful of competitors have also emerged. Soundhound is one such company, which boasts that it can identify a song simply by singing or humming it… with mixed results. There are also song identifiers built into voice apps like Google Assistant, which work very similarly to Shazam's system.
