Google says that users will not need to be pitch-perfect to find the results but instead will show the most likely options based on the tune. It will then provide the song title, artists, and the option to open the song in other apps such as Google-owned YouTube.
Upon testing, Google was able to provide results for the Bee Gee’s Staying Alive and Drake’s Hotline Bling but struggled to identify more challenging songs like Peter Gabriel’s Solsbury Hill.
When you hum a melody into Search, our machine learning models transform the audio into a number-based sequence representing the song’s melody. Our models are trained to identify songs based on a variety of sources, including humans singing, whistling, or humming, as well as studio recordings”, Krishna Kumar, Senior Product Manager of Google Search, explained in a blog post.
“The algorithms also take away all the other details, like accompanying instruments and the voice’s timbre and tone. What we’re left with is the song’s number-based sequence or the fingerprint.”
Google is also improving its ability to understand misspelled words, and says that one in every 10 queries are misspelled.
The new spelling algorithm uses a deep neural net to improve its recognition, a change which “makes a greater improvement to spelling than all of our improvements over the last five years”, Google says.
It recently updated its Android operating system so that it will send users notifications when their phone hears certain household sounds.
The search giant also released its new Nest Audio smart speaker; the update brings a complete redesign and puts it in a new shape.