Our Goals

What we're interested in is the analysis of music video (or opera) structure. The primal steps we might adopt is as follows:

• Audio

1. Separate the non-music (speech, silence) and music (pure-music, vocal) parts.
2. Extract the vocal in the music parts got from step 1.
3. Identifying roles by voice.

• Video

1. Face (role) detection
2. Identifying roles by face (costume) recognition

After getting the clues from audio and video, we would try to analysize the social relationship between the roles

• Social Relationship

1. Who is the leading role
2. How much relevence is between any pair of two roles

If those goals above can be achieved succesfully, some application can be done:

• Application

1. Give users only the fragments of those actors/actresses he/she is interested in
2. The graphics of roles relationship
3. A simple script. I mean the program can automatically make marks for different roles, this make the audience easier to understand the story.