-
Notifications
You must be signed in to change notification settings - Fork 1
/
language_model.js
184 lines (148 loc) · 4.53 KB
/
language_model.js
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
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
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
const fs = require('fs-extra');
const { spawn } = require('child_process');
const { isArray, keys } = require('lodash');
const tmp = require('tmp');
const { tmpName } = require('./util');
// [oov] no longer in words.txt
const OOV_TERM = '<unk>';
tmp.setGracefulCleanup();
class MySet {
constructor(set = []) {
this.set = {};
set.forEach(s => {
this.set[s] = true;
});
}
add(v) {
this.set[v] = true;
}
update(set) {
const items = set.set || set;
items.forEach(s => {
this.set[s] = true;
});
}
}
const setdefault = (obj, key) => {
if (!obj[key]) {
obj[key] = new MySet();
}
return obj[key];
};
const makeBigramLmFst = (wordSequencesIn, options = {}) => {
// Use the given token sequence to make a bigram language model
// in OpenFST plain text format.
// When the "conservative" flag is set, an [oov] is interleaved
// between successive words.
// When the "disfluency" flag is set, a small set of disfluencies is
// interleaved between successive words
// `Word sequence` is a list of lists, each valid as a start
let wordSequences = wordSequencesIn;
if (wordSequences.length === 0 || !isArray(wordSequences[0])) {
wordSequences = [wordSequences];
}
const conservative = options.conservative || false;
const disfluency = options.disfluency || false;
const disfluencies = options.disfluencies || [];
const bigrams = { [OOV_TERM]: new MySet([OOV_TERM]) };
wordSequences.forEach(wordSequence => {
if (wordSequence.length === 0) {
return;
}
let prevWord = wordSequence[0];
bigrams[OOV_TERM].add(prevWord); // valid start (?)
if (disfluency) {
bigrams[OOV_TERM].update(disfluencies);
disfluencies.forEach(dis => {
setdefault(bigrams, dis).add(prevWord);
bigrams[dis].add(OOV_TERM);
});
}
wordSequence.slice(1).forEach(word => {
setdefault(bigrams, prevWord).add(word);
if (conservative) {
bigrams[prevWord].add(OOV_TERM);
}
if (disfluency) {
bigrams[prevWord].update(disfluencies);
disfluencies.forEach(dis => {
bigrams[dis].add(word);
});
}
prevWord = word;
});
// ...valid end
setdefault(bigrams, prevWord).add(OOV_TERM);
});
const nodeIds = {};
function getNodeId(word) {
const nodeId = nodeIds[word] || keys(nodeIds).length + 1;
nodeIds[word] = nodeId;
return nodeId;
}
let output = '';
keys(bigrams).sort().forEach(fromWord => {
const fromId = getNodeId(fromWord);
const successors = keys(bigrams[fromWord].set);
let weight;
if (successors.length > 0) {
weight = -Math.log(1.0 / successors.length);
} else {
weight = 0;
}
successors.sort().forEach(toWord => {
const toId = getNodeId(toWord);
output += `${fromId} ${toId} ${toWord} ${toWord} ${weight}`;
output += '\n';
});
});
output += `${keys(nodeIds).length} 0\n`;
return output;
};
const makeBigramLanguageModel = async (kaldiSeq, resources, options = {}) => {
// Generates a language model to fit the text.
// Returns the filename of the generated language model FST.
// The caller is resposible for removing the generated file.
// `proto_langdir` is a path to a directory containing prototype model data
// `kaldi_seq` is a list of words within kaldi's vocabulary.
const { protoLangDir } = resources;
const MKGRAPH_PATH = await resources.getBinary('m3');
// Generate a textual FST
const txtFst = makeBigramLmFst(kaldiSeq, options);
const txtFstFile = await tmpName();
return fs.writeFile(txtFstFile, txtFst)
.then(async () => {
const hclgFilename = (await tmpName()) + '_HCLG.fst';
const args = [protoLangDir, txtFstFile, hclgFilename];
const p = spawn(MKGRAPH_PATH, args, { stdio: 'ignore' });
return new Promise((resolve, reject) => {
let rejected = false;
p.on('error', err => {
if (!rejected) {
rejected = true;
reject(new Error(`m3 failed with ${err.message}`));
}
});
p.on('exit', code => {
if (code) {
if (!rejected) {
rejected = true;
reject(new Error(`m3 failed with ${code}`));
}
} else {
resolve();
}
});
}).catch(err => {
fs.unlink(txtFstFile);
throw err;
}).then(() => {
fs.unlink(txtFstFile);
return hclgFilename;
});
});
};
module.exports = {
makeBigramLmFst,
makeBigramLanguageModel,
};