File size: 5,142 Bytes
17e27a2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import re
import tempfile
import torch
import sys
import gradio as gr

from huggingface_hub import hf_hub_download

# Setup TTS env
if "vits" not in sys.path:
    sys.path.append("vits")

from vits import commons, utils
from vits.models import SynthesizerTrn


TTS_LANGUAGES = {}
with open(f"data/tts/all_langs.tsv") as f:
    for line in f:
        iso, name = line.split(" ", 1)
        TTS_LANGUAGES[iso] = name


class TextMapper(object):
    def __init__(self, vocab_file):
        self.symbols = [
            x.replace("\n", "") for x in open(vocab_file, encoding="utf-8").readlines()
        ]
        self.SPACE_ID = self.symbols.index(" ")
        self._symbol_to_id = {s: i for i, s in enumerate(self.symbols)}
        self._id_to_symbol = {i: s for i, s in enumerate(self.symbols)}

    def text_to_sequence(self, text, cleaner_names):
        """Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
        Args:
        text: string to convert to a sequence
        cleaner_names: names of the cleaner functions to run the text through
        Returns:
        List of integers corresponding to the symbols in the text
        """
        sequence = []
        clean_text = text.strip()
        for symbol in clean_text:
            symbol_id = self._symbol_to_id[symbol]
            sequence += [symbol_id]
        return sequence

    def uromanize(self, text, uroman_pl):
        iso = "xxx"
        with tempfile.NamedTemporaryFile() as tf, tempfile.NamedTemporaryFile() as tf2:
            with open(tf.name, "w") as f:
                f.write("\n".join([text]))
            cmd = f"perl " + uroman_pl
            cmd += f" -l {iso} "
            cmd += f" < {tf.name} > {tf2.name}"
            os.system(cmd)
            outtexts = []
            with open(tf2.name) as f:
                for line in f:
                    line = re.sub(r"\s+", " ", line).strip()
                    outtexts.append(line)
            outtext = outtexts[0]
        return outtext

    def get_text(self, text, hps):
        text_norm = self.text_to_sequence(text, hps.data.text_cleaners)
        if hps.data.add_blank:
            text_norm = commons.intersperse(text_norm, 0)
        text_norm = torch.LongTensor(text_norm)
        return text_norm

    def filter_oov(self, text, lang=None):
        text = self.preprocess_char(text, lang=lang)
        val_chars = self._symbol_to_id
        txt_filt = "".join(list(filter(lambda x: x in val_chars, text)))
        return txt_filt

    def preprocess_char(self, text, lang=None):
        """
        Special treatement of characters in certain languages
        """
        if lang == "ron":
            text = text.replace("ț", "ţ")
            print(f"{lang} (ț -> ţ): {text}")
        return text


def synthesize(text,speed,lang):
    #lang = "spa"
    #speed =1
    if speed is None:
        speed = 1.0

    lang_code = lang.split()[0].strip()

    vocab_file = hf_hub_download(
        repo_id="facebook/mms-tts",
        filename="vocab.txt",
        subfolder=f"models/{lang_code}",
    )
    config_file = hf_hub_download(
        repo_id="facebook/mms-tts",
        filename="config.json",
        subfolder=f"models/{lang_code}",
    )
    g_pth = hf_hub_download(
        repo_id="facebook/mms-tts",
        filename="G_100000.pth",
        subfolder=f"models/{lang_code}",
    )

    if torch.cuda.is_available():
        device = torch.device("cuda")
    elif (
        hasattr(torch.backends, "mps")
        and torch.backends.mps.is_available()
        and torch.backends.mps.is_built()
    ):
        device = torch.device("mps")
    else:
        device = torch.device("cpu")

    print(f"Run inference with {device}")

    assert os.path.isfile(config_file), f"{config_file} doesn't exist"
    hps = utils.get_hparams_from_file(config_file)
    text_mapper = TextMapper(vocab_file)
    net_g = SynthesizerTrn(
        len(text_mapper.symbols),
        hps.data.filter_length // 2 + 1,
        hps.train.segment_size // hps.data.hop_length,
        **hps.model,
    )
    net_g.to(device)
    _ = net_g.eval()

    _ = utils.load_checkpoint(g_pth, net_g, None)

    is_uroman = hps.data.training_files.split(".")[-1] == "uroman"

    if is_uroman:
        uroman_dir = "uroman"
        assert os.path.exists(uroman_dir)
        uroman_pl = os.path.join(uroman_dir, "bin", "uroman.pl")
        text = text_mapper.uromanize(text, uroman_pl)

    text = text.lower()
    text = text_mapper.filter_oov(text, lang=lang)
    stn_tst = text_mapper.get_text(text, hps)
    with torch.no_grad():
        x_tst = stn_tst.unsqueeze(0).to(device)
        x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).to(device)
        hyp = (
            net_g.infer(
                x_tst,
                x_tst_lengths,
                noise_scale=0.667,
                noise_scale_w=0.8,
                length_scale=1.0 / speed,
            )[0][0, 0]
            .cpu()
            .float()
            .numpy()
        )

    return hps.data.sampling_rate,hyp
    #return gr.Audio.update(value=(hps.data.sampling_rate, hyp)), text