Audio results

Generating Emotionally Relevant Musical Scores for Audio Stories

Steve Rubin, Maneesh Agrawala — UC Berkeley

UIST 2014

The following are a set of musical scores that our system generated. Each example used emotion labels from one of our three approaches: hand-labeling, crowd-labeling, and automatic labeling. Clicking an example with play it and show a visualization of the musical score. Be sure to see the video example at the bottom of the page.

Note: this page requires javascript and in some cases, Flash. Please use Chrome or Safari if possible.

Results used in our crowd listening evaluation


This example is a longer story. We increased the music segment duration minimum to 40 seconds and maximum to 180 seconds in the constraints in order to score the long sections without emotion changes that might appear in longer stories.

In addition to the audio results, we include a sample result for a visual children's story. Watch the video below. This uses hand labeling and two music tracks: "Time Machine" and "The Bathtub". Original YouTube link for the story.