15-494/694 Cognitive Robotics: SpeechNode Assignment

In this assignment you are going to add new functionality to Tekkotsu's speech node. We will use this functionality when we develop our Tic-Tac-Toe playing demo.

Understanding the SpeechNode

  1. You should have already read Tekkotsu and Software Engineering so you know how to browse the online documentation.

  2. Read the online documentation for the SpeechNode class.

  3. There are three ways to use a SpeechNode. We'll look at the most common ones here. First, if the text is a constant, you can supply it in the constructor call when you write your state machine:
    SpeechNode("I'm going for a walk") =C=>
       WalkForward(0.5) =C=>
         SpeechNode("That was fun!")

    Second, if you need to dynamically compute the text to be spoken instead of using a constant, you can make your own subclass of SpeechNode and write text to the member variable textstream:

    $nodeclass Report : SpeechNode : doStart {
       textstream << "I saw " << camShS.allShapes().size() << " shapes.";
    }

    ...

    $statemachine{
       LookForStuff =C=> Report
    }

  4. The SpeechNode listens for an audioEGID event with a specific source ID to know when the speech it generated has finished playing. Then it posts a completion event.

  5. Read the source file SpeechNode.h to understand how this class is implemented.

  6. Note that the SpeechNode constructor takes two string arguments: the name of the node, and (optionally) the text to be spoken. When we make instances of SpeechNode in the state machine language, the name for the node is inserted automatically by the state machine compiler, so we never supply that first argument ourselves.

New Functionality

We want to extend the SpeechNode class to provide an option to select at random from a collection of phrases, in order to give the robot's responses more variety and spontaneity. To implement this option we're going to create a new constructor that takes three arguments:
  • Argument 1: const std::string& nodename
  • Argument 2: const std::string& utterances
  • Argument 3: int repetitions
The utterance argument can take two forms. In the simplest case it's a list of phrases separated by newline characters. In the second case it's the name of a file containing a list of phrases separated by newlines. We will assume that the string is a filename if it contains no newlines and at least one period.

Each time the SpeechNode is entered it picks a new phrase. The repetitions argument controls how phrases are selected. If its value is -1 then the SpeechNode cycles through the phrases in the order they are listed, wrapping around when it has run through the entire set.

If the repetitions argument is zero, the SpeechNode picks a phrase at random from the set provided.

If the repetitions argument is a positive number N, the SpeechNode picks a phrase at random except that it will not reuse a phrase that has already been chosen for one of the previous N utterances. Obviously, N must be less than the number of phrases supplied. If it's not, the node should give an error message and set N to zero.

Here is a quick way to test your implementation:

startnode: MySpeechNode("Fine\n Okay\n Good\n Wonderful\n Peachy", 3)
   =TM=> startnode

Every time you type "msg" in the Tekkotsu console, that generates a blank text message event, which will fire the text message transition, causing the state machine to leave and then immediately reenter the SpeechNode. Listen for repetitions of each phrase.

Implementation Notes

  • Copy /usr/local/Tekkotsu/Behaviors/Nodes/SpeechNode.h to your ~/project directory and rename the file (and the class) to MySpeechNode.

  • To implement file i/o, make your node inherit from the LoadSave class. We will not need any of the fancy serialization features of LoadSave. All we want to do is have it open the file and load the contents into a buffer that we can access. You might want to look at the implementation of other classes that use LoadSave, such as CMPackWalkMC, to see how they work.

  • To learn how to choose randomly from a list, look at the implementation of the RandomTrans class.

  • Your code should fully adhere to Tekkotsu coding standards.

  • You should use doxygen notation for your comments. See the source file for examples.

Hand In

This assignment should be done individually, not in teams.

When you've completed the assignment, email your source file to Dave Touretzky.

Your solution is due on Wednesday, Feb. 3, 2016.

Back to the Cognitive Robotics home page