How to use Google Hum to Search-Machine learning of this Google feature

Most of us are like to hear songs and music. But sometimes, you can’t remember the lyrics of a song after not hearing it for a long time. It happens all the time. But now Google has a way to remember the songs by just humming, or whistling. Let’s see how to use this feature.

Google Hum to Search

This feature is called Hum to Search and the feature is available on all android and iPhones. You can use Google Assistant or Google app for this feature. When you’re using the Google Assistant app, just say, Hey Google, what’s this song?” And then start humming, whistling, or singing. If you use the Google app, tap the microphone icon, and say, “What’s this song?” You can also thump the “search a song” button. And as usual then start humming, whistling, or singing.

Google finds us the exact song we searched for whether it’s lyrics, song, or music video. This feature will also recommend a bunch of songs sometimes when it’s couldn’t find the exact song because everyone can’t hum, sing, or whistle to the pitch. This feature is only available in English for iPhone users and it’s more than 20 languages for Android users currently.

How Google Hum to search works in behind

Google launched the Hum to Search in October 2020. The feature relies on a fully machine-learned system. This technique generates an embedding of a melody from a spectrogram for songs directly. This enables the system to match the hummed or whistled songs to original music instantly without only having pre-recorded hummed recordings or using any complex manual engineering logic.

Machine learning of this feature

When we hum to Google, it develops a sequence for our melody. Google has a database that contains millions of sequences like that. Then Hum to Search system match those sequences and find us the most suitable recording. But in this approach, the system can’t match two sequences directly. Because as I said before, the humming melody could have some pitching issues. And to solve a problem like this, necessarily it’s not used classical algorithm. Because if then, the programme will become more complex and have to change relentlessly. In this scenario, Google used a deep neural network and it could strengthen its system by the experience of hearing songs which have been called music recognition technology.

To succeed in this technology, Google had to generate sequences for songs and maintain a database of sequences for nearly all the songs in the world. Then had to train an AI to match hum or whistle to studio recording. To increase the efficiency of this, engineers of Google have been divided a wave firm into 0.8 seconds portions. Then the AI has been able to match only 0.8 portions of the dual. If you interest to go deep into how the machine learning process and training happens, read this Google AI blog article.

At the moment, Hum to Search feature works really well. We could hope this technology will be advanced more in the future because millions of people using this feature and it would help to improve the experience of the neural network of this system.

At Inventho, we intend to give you more about Science, Technology, Space, Motivational and Inspirational stories in addition to aware you with latest news in the science world. Stay tuned with us by subscribing to our website or Facebook page.