Interactive Frank

1
Interactive Frank

• An interactive audio montage

2
Joe Frank

• Critically acclaimed broadcaster and performance
  artist
• a master of the dark, humorous and absurd
• Shows aired on WFMU and NPR
• Series of fictitious narrative monologues,
  dramas, "real life" phone conversations, and
  often all of the above
• My inspiration for author model

3
Charles Marks

• Old grumpy artist
• Ex-girlfriends grandfather
• Loved to talk about his past
• Could spout off loosely related anecdotes for
  hours on end, with minimal input
• Perfect model for interactivizing Joe Frank

4
Make it Interactive?

• What if the user could guide an Joe Frank-esque
  audio montage in the same way that I could
  guide my conversations with Mr. Marks?

5
Jobs

• Author take input from the user and try to make
  sense of it
• Actors communicate with the author and try to
  say something relevant to the input
• Lenny Wikipedia
• Oswald - LiveJournal
• DJ (and daemon) communicate with author and try
  to play a song from your hard drive that matches
  the overall theme of the input

6
Technical Details

• Google API
• FreeTTS (text-to-speech)
• Not so good, but you can set pitch, WPM, style,
  and its free
• MP3s on your hard drive
• ID3 Tags (blinkenlights) w/help from allmusic.com
• dict.org (especially Moby Thesaurus)
• Wikipedia
• LiveJournal RSS

7
The Interface

• Accepts input from user
• Currently works only with single adjectives
• Eventually, I would like it to accept sentences,
  and do some sort of simple analysis of what words
  are important
• eg surreal, uplifting, horrible, sublime, etc.

8
Meanwhile

• When you start up the program, the DJ begins
  searching your hard drive.
• When it encounters an MP3, it checks the ID3 tags
• DJ keeps track of Artists with lists of moods,
  and songs by that artist
• Moods are expanded using thesaurus

9
Themes

• When you enter a word, a theme is constructed.
• The Theme consults dict.org and retrieves
  synonyms
• When there is more than one valid theme, the
  Author goes to work
• The themes are the primary narrative guide

10
The Author

• The Author analyzes the Themes created by the
  user
• Whenever there is an intersection between two
  themes, we have encountered something important
  (a focus)
• These foci are what is handed to the DJ and the
  Actors as inspiration

nonsensical
superficial
meaningless
foolish
extravagant
lifeless
insubstantial
tedious
11
The Actors

• When the themes have been analyzed, the results
  are fed to the Actors
• The Actors then take turns reading a line
• The have different sources and different ways of
  gathering material

12
Lenny

• Lenny here loves to drop science
• His body of knowledge comes from the WikiPedia
• He finds an article that has something to do with
  the foci and reads a sentence from it

13
Oswald

• Oswald is into drama
• He reads from LiveJournal

14
Modularity

• Authors can be added
• I had to get rid of Fitz (my original character)
  he couldnt get his act together
• Eventually, the authors will have values that
  will determine when they will say a line.

15
The DJ

• The DJ is also taking cues from the Author
• He is going through his list of Artists,
  comparing the moods against the foci that the
  Author has determined
• Once he has found the artist that best matches
  the themes, he begins to play a song.
• The process starts over when the song has ended

16
Is it Narrative?

• No
• The result is a disjointed series of sentences,
  not nearly as cohesive as I want it to be
• What could make it narrative?
• Themes should be influenced by previous input
• Use n-grams
• GoogleAPI should allow more than 1000 queries a
  day

17
Then whats the point?

• I thought that the DJ was remarkable effective.
• Each song he chose was very accurate
• The dialogue was often very entertaining
• I could often fill in the blanks of the dialogue
  and come up with a cohesive narrative of my own
• With more work, it could definitely get better