File size: 8,052 Bytes
f23c138
 
cea6632
 
 
 
 
 
c8c05d4
49696ae
 
cea6632
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ed7f208
 
502b0b6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ec841b5
502b0b6
 
 
 
 
 
 
 
 
 
ec841b5
cea6632
502b0b6
 
ec841b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
502b0b6
 
 
 
 
 
cea6632
 
 
 
 
1d606bc
cea6632
 
5468bc2
ec841b5
5468bc2
cea6632
 
 
 
 
 
 
 
502b0b6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cea6632
 
 
 
 
 
404e64b
1d606bc
404e64b
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
import os

import gradio as gr
import numpy as np
import torch

from InferenceInterfaces.Meta_FastSpeech2 import Meta_FastSpeech2

os.system("pip uninstall -y gradio")
os.system("pip install gradio==2.7.5.2")


def float2pcm(sig, dtype='int16'):
    """
    https://gist.github.com/HudsonHuang/fbdf8e9af7993fe2a91620d3fb86a182
    """
    sig = np.asarray(sig)
    if sig.dtype.kind != 'f':
        raise TypeError("'sig' must be a float array")
    dtype = np.dtype(dtype)
    if dtype.kind not in 'iu':
        raise TypeError("'dtype' must be an integer type")
    i = np.iinfo(dtype)
    abs_max = 2 ** (i.bits - 1)
    offset = i.min + abs_max
    return (sig * abs_max + offset).clip(i.min, i.max).astype(dtype)


class TTS_Interface:

    def __init__(self):
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        self.model = Meta_FastSpeech2(device=self.device)
        self.current_speaker = "English Speaker's Voice"
        self.current_language = "English"
        self.language_id_lookup = {
            "English"   : "en",
            "German"    : "de",
            "Greek"     : "el",
            "Spanish"   : "es",
            "Finnish"   : "fi",
            "Russian"   : "ru",
            "Hungarian" : "hu",
            "Dutch"     : "nl",
            "French"    : "fr",
            'Polish'    : "pl",
            'Portuguese': "pt",
            'Italian'   : "it",
            }
        self.speaker_path_lookup = {
            "English Speaker's Voice"   : "reference_audios/english.wav",
            "German Speaker's Voice"    : "reference_audios/german.wav",
            "Greek Speaker's Voice"     : "reference_audios/greek.wav",
            "Spanish Speaker's Voice"   : "reference_audios/spanish.wav",
            "Finnish Speaker's Voice"   : "reference_audios/finnish.wav",
            "Russian Speaker's Voice"   : "reference_audios/russian.wav",
            "Hungarian Speaker's Voice" : "reference_audios/hungarian.wav",
            "Dutch Speaker's Voice"     : "reference_audios/dutch.wav",
            "French Speaker's Voice"    : "reference_audios/french.wav",
            "Polish Speaker's Voice"    : "reference_audios/polish.flac",
            "Portuguese Speaker's Voice": "reference_audios/portuguese.flac",
            "Italian Speaker's Voice"   : "reference_audios/italian.flac",
            }

    def read(self, prompt, language, speaker):
        if len(prompt) > 2000:
            if language == "English":
                prompt = "Your input was too long. Please try either a shorter text or split it into several parts."
            elif language == "German":
                prompt = "Deine Eingabe war zu lang. Bitte versuche es entweder mit einem kürzeren Text oder teile ihn in mehrere Teile auf."
            elif language == "Greek":
                prompt = "Η εισήγησή σας ήταν πολύ μεγάλη. Παρακαλώ δοκιμάστε είτε ένα μικρότερο κείμενο είτε χωρίστε το σε διάφορα μέρη."
            elif language == "Spanish":
                prompt = "Su entrada es demasiado larga. Por favor, intente un texto más corto o divídalo en varias partes."
            elif language == "Finnish":
                prompt = "Vastauksesi oli liian pitkä. Kokeile joko lyhyempää tekstiä tai jaa se useampaan osaan."
            elif language == "Russian":
                prompt = "Ваш текст слишком длинный. Пожалуйста, попробуйте либо сократить текст, либо разделить его на несколько частей."
            elif language == "Hungarian":
                prompt = "Túl hosszú volt a bevitele. Kérjük, próbáljon meg rövidebb szöveget írni, vagy ossza több részre."
            elif language == "Dutch":
                prompt = "Uw input was te lang. Probeer een kortere tekst of splits het in verschillende delen."
            elif language == "French":
                prompt = "Votre saisie était trop longue. Veuillez essayer un texte plus court ou le diviser en plusieurs parties."
            elif language == 'Polish':
                prompt = "Twój wpis był zbyt długi. Spróbuj skrócić tekst lub podzielić go na kilka części."
            elif language == 'Portuguese':
                prompt = "O seu contributo foi demasiado longo. Por favor, tente um texto mais curto ou divida-o em várias partes."
            elif language == 'Italian':
                prompt = "Il tuo input era troppo lungo. Per favore, prova un testo più corto o dividilo in più parti."

        if self.current_language != language:
            self.model.set_language(self.language_id_lookup[language])
            self.current_language = language
        if self.current_speaker != speaker:
            self.model.set_utterance_embedding(self.speaker_path_lookup[speaker])
            self.current_speaker = speaker
        wav = self.model(prompt)
        return 48000, float2pcm(wav.cpu().numpy())


meta_model = TTS_Interface()
article = "<p style='text-align: left'>This is still a work in progress, models will be exchanged for better ones as soon as they are done. All of those languages are spoken by a single model. Speakers can be transferred across languages. More languages will be added soon.</p><p style='text-align: center'><a href='https://github.com/DigitalPhonetics/IMS-Toucan' target='_blank'>Click here to learn more about the IMS Toucan Speech Synthesis Toolkit</a></p>"

iface = gr.Interface(fn=meta_model.read,
                     inputs=[gr.inputs.Textbox(lines=2,
                                               placeholder="write what you want the synthesis to read here... (2000 character maximum)",
                                               label=" "),
                             gr.inputs.Dropdown(['English',
                                                 'German',
                                                 'Greek',
                                                 'Spanish',
                                                 'Finnish',
                                                 'Russian',
                                                 'Hungarian',
                                                 'Dutch',
                                                 'French',
                                                 'Polish',
                                                 'Portuguese',
                                                 'Italian'], type="value", default='English', label="Language Selection"),
                             gr.inputs.Dropdown(["English Speaker's Voice",
                                                 "German Speaker's Voice",
                                                 "Greek Speaker's Voice",
                                                 "Spanish Speaker's Voice",
                                                 "Finnish Speaker's Voice",
                                                 "Russian Speaker's Voice",
                                                 "Hungarian Speaker's Voice",
                                                 "Dutch Speaker's Voice",
                                                 "French Speaker's Voice",
                                                 "Polish Speaker's Voice",
                                                 "Portuguese Speaker's Voice",
                                                 "Italian Speaker's Voice"], type="value", default="English Speaker's Voice", label="Speaker Selection")],
                     outputs=gr.outputs.Audio(type="numpy", label=None),
                     layout="vertical",
                     title="IMS Toucan Multilingual Multispeaker Demo",
                     thumbnail="Utility/toucan.png",
                     theme="default",
                     allow_flagging="never",
                     allow_screenshot=False,
                     article=article)
iface.launch(enable_queue=True)