Back in October, we highlighted a feature, i.e. Hum to Search that Google has added to its search tools. This would allow users to search for any song, simply just by humming the tune of it.
You may agree or disagree on this front, but this feature is quite a handy and nifty feature that one would like to have in the world of such smart technologies.
Personally, there have been many occasions where I have walked into a coffee shop and there is a catchy tune playing. But for some reason, I do not know from which album the song is or what is the name of the song.
I know this has been a situation that many of you may have been also into. Besides, it could also come in handy for those who could remember the lyrics of the song, but remember the tune of it.
Using this Google Hum to Search feature, users can just hum the tune of the song and Google will try to match and identify the tune and show you the results.
And now Google in official AI blog has explained how this hum to search feature actually works. FYI, Pixel phones 2017's Now Playing feature, could also recognize songs laying in the background without needing an internet connection.
A number-based sequence representing the song's melody is generated when you hum to search
This new hum to search feature is based on the same Now Playing technology of Pixel phones. However, the hum to search feature is faster as it is now server-based.
Moreover, explaining things further Now Playing recognized songs based on recorded audio. On the other hand, when using hum to search, Google generates a number-based sequence.
And this number-based sequence represents the song's melody. Apparently, this ignores the instruments and the quality of the voice. Now, this is compared to the melody, and hum to search shows the best matching percentage results.
Google says that the current database that the hum to search feature can look into for songs is 500,000 songs. But Google defends this lesser number, saying that it is because it has a high level of accuracy.
Comparing hum to search with Sound Search, the latter can search through 100 million songs. Which is way huge of a database. Good thing is that Google will be continuously updating its database with more songs.
So, instead of bitching about the fewer number of song database of hum to search, we should actually appreciate Google for having such a feature. Because in the end, getting to know information about forgotten songs is what matters to the end-user.
For more detailed info on how performed the training of its AI and machine learning process, you can check out the official blog here.