Google tells you the title of the song you are thinking about

Google Song

Google tells us what is the title of the song we are humming or just whistling.

I have in mind a song that says… “da daaaa da daaaa daaaa naa naa naa naa naa ooohhhh yeah“?
And the music goes: “uhmmm uhmmm uhmmm daaaa“?

Now just hum or whistle the tune in your head and Google will search and find it, all thanks to a new feature available with the latest version of the Google app.
You don’t need lyrics, neither the artist’s name, nor the perfect intonation.

How Google searches for the song

On your mobile device, open the latest version of the Google app or find your Google Search widget, tap on the microphone icon and say “what is this song?” or click on the “Search a song” button.
Then start humming for 10-15 seconds.

On Google Assistant, it’s just as simple. Say “Hey Google, what’s this song?” and then hum the tune.

At this point, Google, based on its automatic learning algorithm, provides its answer: the list of songs with more affinity to the one you have heard.
The answers are different versions of the song interpreted by different artists, or by the same artist with different arrangements.
For each song, then, Google gives a link to watch the video (on YouTube, of course), listen to it with your favourite music, or read the lyrics.

The function is now available in English on iOs devices and in 20 languages on Android (including Italian), but Google’s intention is to add new languages, and further improve the search mechanism, exploiting the infinite possibilities of its machine learning models, those that transform our audio into a sequence of numbers representing the melody of the song to give an answer to our question.

The machine learning model

But how do you recognise the song?
The melody of a song is like its fingerprint: each one has its own unique identity.
Google has developed machine learning models that can combine whistling or singing with the right “fingerprint“.

When you hum a melody in Search, machine learning models transform the audio into a numerical sequence that represents the melody of the song and identify songs based on a variety of sources, including human singing or whistling.
The algorithms also eliminate all other details, such as the accompanying instruments and the timbre and tone of the voice.
What we are left with is the numerical sequence of the song, or the fingerprint.

These sequences are compared with thousands of songs from all over the world, identifying possible matches in real time.

For all references: