public static void main(String[] args) throws IOException {
// Parse command line flags and arguments
final Map<String, String> argMap = CommandLineUtils.simpleCommandLineParser(args);
// Set up default parameters and settings
String basePath = ".";
String model = "baseline";
boolean verbose = false;
// Update defaults using command line specifications
// The path to the assignment data
if (argMap.containsKey("-path")) {
basePath = argMap.get("-path");
}
System.out.println("Using base path: " + basePath);
// A string descriptor of the model to use
if (argMap.containsKey("-model")) {
model = argMap.get("-model");
}
System.out.println("Using model: " + model);
// Whether or not to print the individual speech errors.
if (argMap.containsKey("-verbose")) {
verbose = true;
}
if (argMap.containsKey("-quiet")) {
verbose = false;
}
// Read in all the assignment data
final String trainingSentencesFile = "/treebank-sentences-spoken-train.txt";
final String speechNBestListsPath = "/wsj_n_bst";
final Collection<List<String>> trainingSentenceCollection =
SentenceCollection.Reader.readSentenceCollection(basePath + trainingSentencesFile);
final Set<String> trainingVocabulary = extractVocabulary(trainingSentenceCollection);
final List<SpeechNBestList> speechNBestLists =
SpeechNBestList.Reader.readSpeechNBestLists(
basePath + speechNBestListsPath, trainingVocabulary);
// String validationSentencesFile =
// "/treebank-sentences-spoken-validate.txt";
// Collection<List<String>> validationSentenceCollection =
// SentenceCollection.Reader.readSentenceCollection(basePath +
// validationSentencesFile);
// String testSentencesFile = "/treebank-sentences-spoken-test.txt";
// Collection<List<String>> testSentenceCollection =
// SentenceCollection.Reader.readSentenceCollection(basePath +
// testSentencesFile);
// Build the language model
LanguageModel languageModel = null;
if (model.equalsIgnoreCase("baseline")) {
languageModel = new EmpiricalUnigramLanguageModel(trainingSentenceCollection);
} else if (model.equalsIgnoreCase("sri")) {
languageModel = new SriLanguageModel(argMap.get("-sri"));
} else if (model.equalsIgnoreCase("bigram")) {
languageModel = new EmpiricalBigramLanguageModel(trainingSentenceCollection);
} else if (model.equalsIgnoreCase("trigram")) {
languageModel = new EmpiricalTrigramLanguageModel(trainingSentenceCollection);
} else if (model.equalsIgnoreCase("katz-bigram")) {
languageModel = new KatzBigramLanguageModel(trainingSentenceCollection);
} else if (model.equalsIgnoreCase("katz-bigram-pp")) {
languageModel = new KatzPPBigramLanguageModel(trainingSentenceCollection);
} else if (model.equalsIgnoreCase("katz-trigram")) {
throw new IllegalStateException(
"Katz trigram model not fully implemented -- remove exception and uncomment next line if implemented");
// languageModel = new KatzTrigramLanguageModel(
// trainingSentenceCollection);
} else {
throw new RuntimeException("Unknown model descriptor: " + model);
}
// Evaluate the language model
// final double wsjPerplexity = calculatePerplexity(languageModel,
// testSentenceCollection);
final double hubPerplexity =
calculatePerplexity(languageModel, extractCorrectSentenceList(speechNBestLists));
// System.out.println("WSJ Perplexity: " + wsjPerplexity);
System.out.println("HUB Perplexity: " + hubPerplexity);
System.out.println("WER Baselines:");
System.out.println(" Best Path: " + calculateWordErrorRateLowerBound(speechNBestLists));
System.out.println(" Worst Path: " + calculateWordErrorRateUpperBound(speechNBestLists));
System.out.println(" Avg Path: " + calculateWordErrorRateRandomChoice(speechNBestLists));
final double wordErrorRate = calculateWordErrorRate(languageModel, speechNBestLists, verbose);
System.out.println("HUB Word Error Rate: " + wordErrorRate);
System.out.println("Generated Sentences:");
for (int i = 0; i < 10; i++) System.out.println(" " + languageModel.generateSentence());
}
