natural language daemon for the media center.
My winter break project (12/28/13 - 1/10/14) by @jitl
What it does
teeveed is a media center daeon you control with your voice. Right now, you talk to it with your smartphone's web browser. You tell teeveed to play a movie or an episode of a show you want to see that you have on your share, and it opens VLC for you.
Behind the scenes, teeveed runs an indexer to keep finding files fast. It runs a webserver so you can communicate with it, and it has a nice pleasent heads-up display, so you can see what it's doing.
Sequelfor our ruby-hipster ORM. DataMapper does too much work that keeps us from raw SQL, which we need to implement trigram-based text matching
Use the JavaFX entry point. Right now teeveed boots from
org.jruby.JarBootstrapMain. We need to use a JavaFX main to use JavaFX bundling tools and eliminate dependence on
mvn com.zenjava:javafx-maven-plugin:2.0:fix-classpathfor deployment
Switch from JRuby-based JIT compiling to AOT compiling. teeveed boots really, really slowley, and that's because there's a lot of JIT happening when the jar is loaded. Maybe things would go faster if we pre-compiled teeveed (and its gems).
- Java JDK 1.7
- Maven 2+
- JRuby 1.7.x
- PostgreSQL 9.1+ with the
pg_trgmextension! pg_trgm is one of the ways we find similar-text matches
teeveed uses gem-maven-plugin instead of
Bundler to manage
pom.xml for more information.
git clone https://github.com/justjake/teeveedor similar to get the source
mvn initializewill download and install the required rubygems in
source env.shwill correct your
$RUBY_LIBenvironment variables to point to the gems installed in
teeveed/target/rubygemsand the teeveed sources.
- Hack away. run
teeveed(alias provided by env.sh) to start the daemon, or
clito start the Pry command line.
If things feel too slow,
You should get drip in your ~/bin,
source drip.env.sh before launching any of this ruby junk.
Drip'll put the spring back in your step!
- Java JDK 1.7
- Maven 2+ (for HUD)
- a Postgres database
The only reason we need Maven and the JDK is because (right now) teeveed isn't packaged as a JavaFX runtime app. Because the JavaFX runtime does some interesting native loading things, the simplest way is just to move the JavaFX libraries into the implicit system classpath.
I will get around to fixing this, but it's low priority for my own machines.
(sudo) mvn com.zenjava:javafx-maven-plugin:2.0:fix-classpath
This moves JavaFX onto the classpath so the JavaFX libraries are always loaded. THIS STEP IS ONLY REQUIRED ON USER-INTERFACE MACHINES
make a directory for all of the teeveed resources. I like
example.teeveed.conf.rbinto your teeveed folder.
~/binor somewhere else on your path, and modify
$TEEVEED_HOMEin it to point to your teeveed folder.
Create a Postgres database and user for teeveed. Make sure your postgres instance is accessable over TCP (unencrypted)
example.teeveed.conf.rbso that it matches your setup. Important things to change:
databasecommand to match the setup of your Postgres database
librarycommand's path should be '/path/to/your/media'
- You'll need to insert your Wit.ai token for the
- Keep a fast index of all the media on a share for quick search (DONE)
- index needs to respond to natural-language searches (DONE)
- few spelling mistakes, mostly word transpositions (TODO: switch to trigram)
- deleted items should be pruned regularly (DONE)
- delegte music indexing and search to Spotify (TODO)
The index is stored in a PostgeSQL database, which gets us nice
full-text-search for free. Indexing is performed by crawling each
library section on a specified schedule, and adding new items, and
removing old items based on a
At some point we should use the following algorithm when returning search results, so that we never return a file that doesn't exist:
# Array<Media> returned in ranked order from best to worst match results = root.search(:title => "the good the bad the ugly", :sense => :watch) results.each do |hit| path = Pathname.new(root.path) + hit.relative_path if not path.exist? hit.delete! next end path = path.realpath.to_s return Result.new(hit, path) end
Natural Language Interface
User story: you sit down on the couch and say, "Ok tv. do we have the new Game of Thrones?"
User story: you want to quickly start music playing while in the kitchen. "Ok tv. Play songs by Royksopp"
We use Wit.ai to interpret user intents from plain
text. Intent behavior is defined in
which performs dispatch based on intent type.
Visual indicators (DONE)
teeveed uses a simple JavaFX window to display user intents and matched entities on-screen. All intents are automatically displayed before they are handled, and the results of the handler (anything with a :friendly_name) function are displayed when the handler completes.