Page 182 - The Definitive Guide to Building Java Robots

Preston_5564C05.fm Page 163 Tuesday, September 20, 2005 5:13 AM

CHAPTER 5 ■ SPEECH 163

} catch (Exception e) {
//print error and exit.
e.printStackTrace();
System.exit(1);
}
}
}

Section Summary

In this section, I showed two examples of using speech recognition, both of which implemented
the JRecognizer interface. The first used Sphinx-4 for Java to demonstrate command and control.
The second used JNI and the Microsoft speech recognition engine to give an example of
continuous dictation.
Before introducing the first example, I gave a brief overview of the Java Speech Grammar
Format (JSGF). This is the format of what words the recognizer needs to understand.
The classes discussed were
• SphinxSR: This is the Java recognizer using command and control.

• MicrosoftSR: This is the continuous recognizer using JNI and Microsoft speech recognition
engine. (You must train this engine prior to use.)

5.3 Chapter Summary

In this chapter, I talked about how to get your robot to talk and recognize what you’re saying.
I also showed you how to use speech technology written in other languages, and I gave a simple
introduction to the Java Native Interface (JNI).
The JNI example, TempConvert, does a simple temperature conversion using a C++
project and class to perform the calculation. You should be able to follow the step-by-step
example to create your own JNI classes.
In section 5.1, I introduced the JVoice interface to standardize implementation of the three
text-to-speech classes. The first, JavaVoice, uses the Java Speech API (JSAPI) to produce speech
from text. The second, FreeTTSVoice, uses the FreeTTS speech synthesis libraries to produce
speech from text. The third, MicrosoftVoice, uses JNI to connect the speech synthesizer engine
built in to my system after installing the Microsoft Speech API (MSAPI). Also in section 5.1, I
showed a sample class that tests the voice quality of all three speech synthesis engines.
In section 5.2, I introduced the JRecognizer interface, used for standardizing implementa-
tions of the following two speech recognition classes. The first recognizer, SphinxSR, uses a
grammar file, a configuration file, and the synthesizer from the Sphinx-4 project. The second
recognizer uses JNI to connect to the Microsoft speech recognition engine. Also in section 5.2,
I wrote a sample class that echoes what’s heard in continuous dictation.
Personally, I like MicrosoftVoice for sound quality and SphinxSR for recognition. I’ll use
these in Chapter 9.
Now that our PC can speak and listen, it’s time to get it to see, which is the topic of the next
chapter.