Package edu.cmu.sphinx.linguist

Examples of edu.cmu.sphinx.linguist.WordSequence

65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
        HashMap<WordSequence, Integer> bigrams = new HashMap<WordSequence, Integer>();
        HashMap<WordSequence, Integer> trigrams = new HashMap<WordSequence, Integer>();
        int wordCount = 0;
       
        if (words.size() > 0) {
            addSequence(unigrams, new WordSequence(words.get(0)));
            wordCount++;
        }
           
        if (words.size() > 1) {
            wordCount++;
            addSequence(unigrams, new WordSequence(words.get(1)));
            addSequence(bigrams, new WordSequence(words.get(0), words.get(1)));
        }

        if (words.size() > 2) {
            addSequence(bigrams, new WordSequence(words.get(1), words.get(2)));
            addSequence(trigrams, new WordSequence(words.get(0), words.get(1), words
                    .get(2)));
        }

        for (int i = 2; i < words.size(); ++i) {
            wordCount++;
            addSequence(unigrams, new WordSequence(words.get(i)));
            addSequence(bigrams, new WordSequence(words.get(i - 1), words.get(i)));
            addSequence(trigrams, new WordSequence(words.get(i - 2),
                                          words.get(i - 1),
                                          words.get(i)));
        }

        float discount = .5f;
        float deflate = 1 - discount;
        Map<WordSequence, Float> uniprobs = new HashMap<WordSequence, Float>();
        for (Map.Entry<WordSequence, Integer> e : unigrams.entrySet()) {
            uniprobs.put(e.getKey(),
                         (float) e.getValue() * deflate / wordCount);
        }

        LogMath lmath = LogMath.getLogMath();
        float logUnigramWeight = lmath.linearToLog(unigramWeight);
        float invLogUnigramWeight = lmath.linearToLog(1 - unigramWeight);
        float logUniformProb = -lmath.linearToLog(uniprobs.size());

        Set<WordSequence> sorted1grams = new TreeSet<WordSequence>(unigrams.keySet());
        Iterator<WordSequence> iter =
                new TreeSet<WordSequence>(bigrams.keySet()).iterator();
        WordSequence ws = iter.hasNext() ? iter.next() : null;
        for (WordSequence unigram : sorted1grams) {
            float p = lmath.linearToLog(uniprobs.get(unigram));
            p += logUnigramWeight;
            p = lmath.addAsLinear(p, logUniformProb + invLogUnigramWeight);
            logProbs.put(unigram, p);

            float sum = 0.f;
            while (ws != null) {
                int cmp = ws.getOldest().compareTo(unigram);
                if (cmp > 0) {
                    break;
                }
                if (cmp == 0) {
                    sum += uniprobs.get(ws.getNewest());
                }
                ws = iter.hasNext() ? iter.next() : null;
            }

            logBackoffs.put(unigram, lmath.linearToLog(discount / (1 - sum)));
        }

        Map<WordSequence, Float> biprobs = new HashMap<WordSequence, Float>();
        for (Map.Entry<WordSequence, Integer> entry : bigrams.entrySet()) {
            int unigramCount = unigrams.get(entry.getKey().getOldest());
            biprobs.put(entry.getKey(),
                        entry.getValue() * deflate / unigramCount);
        }

        Set<WordSequence> sorted2grams = new TreeSet<WordSequence>(bigrams.keySet());
        iter = new TreeSet<WordSequence>(trigrams.keySet()).iterator();
        ws = iter.hasNext() ? iter.next() : null;
        for (WordSequence biword : sorted2grams) {
            logProbs.put(biword, lmath.linearToLog(biprobs.get(biword)));

            float sum = 0.f;
            while (ws != null) {
                int cmp = ws.getOldest().compareTo(biword);
                if (cmp > 0) {
                    break;
                }
                if (cmp == 0) {
                    sum += biprobs.get(ws.getNewest());
                }
                ws = iter.hasNext() ? iter.next() : null;
            }
            logBackoffs.put(biword, lmath.linearToLog(discount / (1 - sum)));
        }
View Full Code Here
320
321
322
323
324
325
326
327
328
329
330
                    if (wordObject == null) {
                        wordObject = Word.UNKNOWN;
                    }
                    wordList.add(wordObject);
                }
                WordSequence wordSequence = new WordSequence(wordList);
                if (tok.hasMoreTokens()) {
                    log10Backoff = Float.parseFloat(tok.nextToken());
                }
                float logProb = logMath.log10ToLog(log10Prob);
                float logBackoff = logMath.log10ToLog(log10Backoff);
View Full Code Here
118
119
120
121
122
123
124
125
                    wordToClassProb = classProbability.getLogProbability();
                }
            }
        }
        float classBasedProbability =
            classLM.getProbability(new WordSequence(classes));
        return classBasedProbability + wordToClassProb;
    }
View Full Code Here
35
36
37
38
39
40
41
42
43
44
45
     *
     * @param t the token to add
     */
    public void add(Token t) {
        stateCount++;
        WordSequence ws = getWordSequence(t);
        WordStats stats = statMap.get(ws);
        if (stats == null) {
            stats = new WordStats(ws);
            statMap.put(ws, stats);
        }
View Full Code Here
39
40
41
42
43
44
45
46
47
48
49
     *
     * @param t the token to add
     */
    public void add(Token t) {
        stateCount++;
        WordSequence ws = getWordSequence(t);
        WordStats stats = statMap.get(ws);
        if (stats == null) {
            stats = new WordStats(ws);
            statMap.put(ws, stats);
        }
View Full Code Here
85
86
87
88
89
90
91
                Word word = wordState.getPronunciation().getWord();
                wordList.add(0, word);
            }
            token = token.getPredecessor();
        }
        return new WordSequence(wordList);
    }
View Full Code Here
TOP

Related Classes of edu.cmu.sphinx.linguist.WordSequence

  • edu.cmu.sphinx.decoder.search.stats.StateHistoryTracker
  • edu.cmu.sphinx.decoder.search.stats.WordTracker
  • edu.cmu.sphinx.linguist.language.classes.ClassBasedLanguageModel
  • edu.cmu.sphinx.linguist.language.grammar.LMGrammar
  • edu.cmu.sphinx.linguist.language.ngram.DynamicTrigramModel
  • edu.cmu.sphinx.linguist.language.ngram.DynamicTrigramModelTest
  • edu.cmu.sphinx.linguist.language.ngram.large.LargeNGramModel
  • edu.cmu.sphinx.linguist.language.ngram.large.LargeNgramTest
  • edu.cmu.sphinx.linguist.language.ngram.SimpleNGramModel
  • edu.cmu.sphinx.linguist.lextree.HMMTree
    • Copyright © 2018 www.massapicom. All rights reserved.
    All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.