The machine learning model used in the experiment was trained on the real voices of four opera singers. Christian Joel (tenor), Frederick Tong (bass), Joanna Gamble (mezzo soprano) and Olivia Doutney (soprano) recorded 16 hours of audio for the experiment. Ingunn Gyda Hrafnkelsdottir and John Holland-Avery provided additional vocals.
But what you hear when you create your very own mini opera using a mouse cursor is actually what the AI determines opera should sound like based on what it's learned from those recordings.
Google says that the Blob Opera will only work in browsers and on machines that rock the latest web technology, and may not work on mobile devices.
To get the blobs blobbing and the gorgeous harmonies sounding, you click on one of the characters and drag it up or down to alter pitch and forward or back to change mouth shape for different vowel sounds. Individual or multiple blobs can be silenced by muting the audio icon, and the eyes follow the mouse cursor around the screen.
Performances can be recorded for playback and sharing, and clicking on the tree icon plops some festive hats on the heads of the blobs and offers their takes on popular Carols.
Project page: Blob Opera