File size: 6,633 Bytes
e4b9a7b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
# -*- coding: utf-8 -*-
# SPDX-FileCopyrightText: 2016-2025 PyThaiNLP Project
# SPDX-FileType: SOURCE
# SPDX-License-Identifier: Apache-2.0
"""
Thai Word-to-Phoneme (Thai W2P)
GitHub : https://github.com/wannaphong/Thai_W2P
"""

from typing import Union

import numpy as np

from pythainlp.corpus import download, get_corpus_path

_GRAPHEMES = list(
    "พจใงต้ืฮแาฐฒฤๅูศฅถฺฎหคสุขเึดฟำฝยลอ็ม"
    + " ณิฑชฉซทรฏฬํัฃวก่ป์ผฆบี๊ธญฌษะไ๋นโภ?"
)
_PHONEMES = list(
    "-พจใงต้ืฮแาฐฒฤูศฅถฺฎหคสุขเึดฟำฝยลอ็ม"
    + " ณิฑชฉซทรํฬฏ–ัฃวก่ปผ์ฆบี๊ธฌญะไษ๋นโภ?"
)

_MODEL_NAME = "thai_w2p"


class _Hparams:
    batch_size = 256
    enc_maxlen = 30 * 2
    dec_maxlen = 40 * 2
    num_epochs = 50 * 2
    hidden_units = 64 * 8
    emb_units = 64 * 4
    graphemes = ["<pad>", "<unk>", "</s>"] + _GRAPHEMES
    phonemes = ["<pad>", "<unk>", "<s>", "</s>"] + _PHONEMES
    lr = 0.001


hp = _Hparams()


def _load_vocab():
    g2idx = {g: idx for idx, g in enumerate(hp.graphemes)}
    idx2g = dict(enumerate(hp.graphemes))

    p2idx = {p: idx for idx, p in enumerate(hp.phonemes)}
    idx2p = dict(enumerate(hp.phonemes))
    # note that g and p mean grapheme and phoneme respectively.
    return g2idx, idx2g, p2idx, idx2p


class Thai_W2P():
    def __init__(self):
        super().__init__()
        self.graphemes = hp.graphemes
        self.phonemes = hp.phonemes
        self.g2idx, self.idx2g, self.p2idx, self.idx2p = _load_vocab()
        self.checkpoint = get_corpus_path(_MODEL_NAME, version="0.2")
        if self.checkpoint is None:
            download(_MODEL_NAME, version="0.2")
            self.checkpoint = get_corpus_path(_MODEL_NAME)
        self._load_variables()

    def _load_variables(self):
        self.variables = np.load(self.checkpoint, allow_pickle=True)
        # (29, 64). (len(graphemes), emb)
        self.enc_emb = self.variables.item().get("encoder.emb.weight")
        # (3*128, 64)
        self.enc_w_ih = self.variables.item().get("encoder.rnn.weight_ih_l0")
        # (3*128, 128)
        self.enc_w_hh = self.variables.item().get("encoder.rnn.weight_hh_l0")
        # (3*128,)
        self.enc_b_ih = self.variables.item().get("encoder.rnn.bias_ih_l0")
        # (3*128,)
        self.enc_b_hh = self.variables.item().get("encoder.rnn.bias_hh_l0")

        # (74, 64). (len(phonemes), emb)
        self.dec_emb = self.variables.item().get("decoder.emb.weight")
        # (3*128, 64)
        self.dec_w_ih = self.variables.item().get("decoder.rnn.weight_ih_l0")
        # (3*128, 128)
        self.dec_w_hh = self.variables.item().get("decoder.rnn.weight_hh_l0")
        # (3*128,)
        self.dec_b_ih = self.variables.item().get("decoder.rnn.bias_ih_l0")
        # (3*128,)
        self.dec_b_hh = self.variables.item().get("decoder.rnn.bias_hh_l0")
        # (74, 128)
        self.fc_w = self.variables.item().get("decoder.fc.weight")
        # (74,)
        self.fc_b = self.variables.item().get("decoder.fc.bias")

    def _sigmoid(self, x):
        return 1 / (1 + np.exp(-x))

    def _grucell(self, x, h, w_ih, w_hh, b_ih, b_hh):
        rzn_ih = np.matmul(x, w_ih.T) + b_ih
        rzn_hh = np.matmul(h, w_hh.T) + b_hh

        rz_ih, n_ih = (
            rzn_ih[:, : rzn_ih.shape[-1] * 2 // 3],
            rzn_ih[:, rzn_ih.shape[-1] * 2 // 3 :],
        )
        rz_hh, n_hh = (
            rzn_hh[:, : rzn_hh.shape[-1] * 2 // 3],
            rzn_hh[:, rzn_hh.shape[-1] * 2 // 3 :],
        )

        rz = self._sigmoid(rz_ih + rz_hh)
        r, z = np.split(rz, 2, -1)

        n = np.tanh(n_ih + r * n_hh)
        h = (1 - z) * n + z * h

        return h

    def _gru(self, x, steps, w_ih, w_hh, b_ih, b_hh, h0=None) -> np.ndarray:
        if h0 is None:
            h0 = np.zeros((x.shape[0], w_hh.shape[1]), np.float32)
        h = h0  # initial hidden state

        outputs = np.zeros((x.shape[0], steps, w_hh.shape[1]), np.float32)
        for t in range(steps):
            h = self._grucell(x[:, t, :], h, w_ih, w_hh, b_ih, b_hh)  # (b, h)
            outputs[:, t, ::] = h

        return outputs

    def _encode(self, word: str) -> np.ndarray:
        chars = list(word) + ["</s>"]
        x = [self.g2idx.get(char, self.g2idx["<unk>"]) for char in chars]
        x = np.take(self.enc_emb, np.expand_dims(x, 0), axis=0)

        return x

    def _short_word(self, word: str) -> Union[str, None]:
        self.word = word
        if self.word.endswith("."):
            self.word = self.word.replace(".", "")
            self.word = "-".join([i + "อ" for i in list(self.word)])
            return self.word
        return None

    def _predict(self, word: str) -> str:
        short_word = self._short_word(word)
        if short_word is not None:
            return short_word

        # encoder
        enc = self._encode(word)
        enc = self._gru(
            enc,
            len(word) + 1,
            self.enc_w_ih,
            self.enc_w_hh,
            self.enc_b_ih,
            self.enc_b_hh,
            h0=np.zeros((1, self.enc_w_hh.shape[-1]), np.float32),
        )
        last_hidden = enc[:, -1, :]

        # decoder
        dec = np.take(self.dec_emb, [2], axis=0)  # 2: <s>
        h = last_hidden

        preds = []
        for _ in range(20):
            h = self._grucell(
                dec,
                h,
                self.dec_w_ih,
                self.dec_w_hh,
                self.dec_b_ih,
                self.dec_b_hh,
            )  # (b, h)
            logits = np.matmul(h, self.fc_w.T) + self.fc_b
            pred = logits.argmax()
            if pred == 3:
                break
            preds.append(pred)
            dec = np.take(self.dec_emb, [pred], axis=0)

        preds = [self.idx2p.get(idx, "<unk>") for idx in preds]

        return preds

    def __call__(self, word: str) -> str:
        if not any(letter in word for letter in self.graphemes):
            pron = [word]
        else:  # predict for oov
            pron = self._predict(word)

        return "".join(pron)


_THAI_W2P = Thai_W2P()


def pronunciate(text: str) -> str:
    """
    Convert a Thai word to its pronunciation in Thai letters.

    Input should be one single word.

    :param str text: Thai text to be pronunciated

    :return: A string of Thai letters indicating
             how the input text should be pronounced.
    """
    return _THAI_W2P(text)