Recall from last year's summer blockbuster hit "Transformers" there was a robot featured named Bumblebee that could not speak like the rest of the robots.
Instead, he used clips of audio that contained the words he wanted to communicate. At the surface, this appears to be a simple task; just search for audio files containing the words you want. This isn't as easy as it seems. In Transformers, Bumblebee would "speak" using audio clips that reflected his mood. If he was energetic and happy, his "speech" clip was energetic and happy. Today's computers are not capable of determining the "mood" of audio. Mood is a subjective emotion that varies from person to person.
If we wanted to ever create a machine like Bumblebee, we need to teach our computers how to determine the mood of songs and audio. One approach to this problem, is to get people to help the computers "compute" mood. I have developed a collaborative, two player game, that lets people listen to an audio clip, and provide the mood labels every second of game play. Read more about the game at http://schubert.ece.drexel.edu/moodswings.
Monday, February 18, 2008
Subscribe to:
Posts (Atom)