Examples of edu.cmu.sphinx.result.Node

edu.cmu.sphinx.result.Node

A node is part of Lattices, representing the theory that a word was spoken over a given period of time. A node also has a set of entering and leaving {@link edu.cmu.sphinx.result.Edge edges}, connecting it to other nodes. One can get and set the beginning and end frames of the word via the getBeginTime and getEndTime methods. When setting these times, the beginning time must be earlier or equal to the end time, otherwise an error will be thrown.

The posterior probability of any word in a word lattice is the probability that the node representing that word occurs on any path through the lattice. It is usually computed as the ratio of the total likelihood scores of all paths through the lattice that pass through the node, to the total likelihood score of all paths through the lattice. Path scores are usually computed using the acoustic likelihoods of the nodes, although language scores can also be incorporated. The posterior probabilities of an entire lattice is usually computed efficiently using the Forward-Backward Algorithm. Refer to the {@link edu.cmu.sphinx.result.Lattice#computeNodePosteriors computeNodePosteriors} method in the Lattice class for details.


      LogMath logMath = LogMath.getLogMath();
      
    Lattice lattice = new Lattice();


    Node a = lattice.addNode("A", "A", 0, 0);
    Node b = lattice.addNode("B", "B", 0, 0);
    Node c = lattice.addNode("C", "C", 0, 0);
    Node d = lattice.addNode("D", "D", 0, 0);


    double acousticAB = 4;
    double acousticAC = 6;
    double acousticCB = 1;
    double acousticBD = 5;
    double acousticCD = 2;


    lattice.setInitialNode(a);
    lattice.setTerminalNode(d);


    lattice.addEdge(a, b, logMath.linearToLog(acousticAB), 0);
    lattice.addEdge(a, c, logMath.linearToLog(acousticAC), 0);
    lattice.addEdge(c, b, logMath.linearToLog(acousticCB), 0);
    lattice.addEdge(b, d, logMath.linearToLog(acousticBD), 0);
    lattice.addEdge(c, d, logMath.linearToLog(acousticCD), 0);


    lattice.computeNodePosteriors(1.0f);
    double pathABD = acousticAB * acousticBD;
    double pathACBD = acousticAC * acousticCB * acousticBD;
    double pathACD = acousticAC * acousticCD;
    double allPaths = pathABD + pathACBD + pathACD;


    double bPosterior = (pathABD + pathACBD) / allPaths;
    double cPosterior = (pathACBD + pathACD) / allPaths;


    double delta = 1e-4;
    Assert.assertEquals (logMath.logToLinear((float) a.getPosterior()), 1.0, delta);
    Assert.assertEquals (logMath.logToLinear((float) b.getPosterior()), bPosterior, delta);
    Assert.assertEquals (logMath.logToLinear((float) c.getPosterior()), cPosterior, delta);
    Assert.assertEquals (logMath.logToLinear((float) d.getPosterior()), 1.0, delta);
  }

Examples of edu.cmu.sphinx.result.Node

Related Classes of edu.cmu.sphinx.result.Node