Google’s New Machine Learning Experiment Lets You Create Your Own Opera Performance

In a wacky fusion of machine learning and classical music, Google has launched its experimental "Blob Opera" - a quartet of animated blobs singing opera music.

According to a blog post from Freya Murray, Google Arts & Culture Program Manager, the machine learning experiment is the brainchild of artist David Li, an interactive artist who has also lent his talents to Cartoon Network's adult-oriented block, Adult Swim. In Blob Opera, four brightly colored blobs start falling on the screen one by one, each with their own voice registers: bass, tenor, mezzo-soprano, and soprano.

Users can drag each blob to sing, with the vertical direction corresponding to pitch and the cursor's horizontal movement adjusting the vowel sound the blob makes. The blobs also grow taller or sway toward the dragging motion direction, with all of their eyes following the cursor - even on standby! Also, as a single blob follows the pitch and the vowel, other blobs instinctively follow the lead. Each blob can also be muted with the speaker button above their heads.

A record button is also found on the lower-left corner of the Blob Opera screen (for desktop users), allowing users to record their own experiments on the four-opera blobs. A pine tree button rests on the lower-right corner for users interested in the blobs' rendition of classical Christmas songs. Snowdrops and Santa hats adorn the blobs. A list of tunes is available for users to choose from: songs include "Jingle Bells" to "The First Noel."

A Tribute to the Original Instrument

In its "About" widget, Blob Opera explains that the experiment "pays tribute to and explores the original musical instrument: the voice."

Li and Google Arts and Culture developed a machine learning model that was then trained based on four actual opera singers' voices. They enlisted the vocal prowesses of Christian Joel (Tenor), Frederick Tong (Bass), Joanna Gamble (Mezzo-Soprano), and Olivia Doutney (Soprano). The vocalists recorded around 16 hours of singing.

While users won't exactly hear these opera singers' voices, their respective recordings trained the machine learning model on what opera singing sounds like. Additionally, Icelandic vocalist Ingunn Gyda Hrafnkelsdottir and British-Iranian baritone John Holland-Avery provided additional vocals to develop the machine learning model.

"What you're hearing isn't real opera singing, but the neural network's understanding of what opera singing sounds like," Li explains in the behind-the-scenes video "Making Blob Opera with David Li."

Google's Holiday Projects

Aside from the vocally-charged Blob Opera and its Christmas playlist, Google Arts & Culture has also prepared interactive activities for other holidays this season. Searching for the Jewish "Hanukkah" or the African-American "Kwanzaa" will include a "Color Your Favorite Artwork" tab in its search results. Each outlined craft can be colored by selecting a color and clicking a layer or a section of the artwork, with Google automatically filling the section. Artworks include the 39c Kwanzaa Stamp published by the United States Postal Service in 2006, a pinback button that celebrates The Seven Principles of Kwanzaa from the Smithsonian National Museum of African American History Culture. The Hanukkah gallery includes Marc Chagall's 1933 artwork, "Solitude," kept in the Tel Aviv Museum of Art.

Check out more news and information on Google in Science Times.