File size: 8,053 Bytes
c77ff04
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cafb434
c77ff04
 
 
 
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
import tempfile
from argparse import Namespace
from pathlib import Path

import gradio as gr
import soundfile as sf
import torch
from matcha.cli import (MATCHA_URLS, VOCODER_URL, assert_model_downloaded,
                        get_device, load_matcha, load_vocoder, process_text,
                        to_waveform)
from matcha.utils.utils import get_user_data_dir, plot_tensor

LOCATION = Path(get_user_data_dir())

args = Namespace(
    cpu=False,
    model="matcha_ljspeech",
    vocoder="hifigan_T2_v1",
    spk=None,
)

MATCHA_TTS_LOC = LOCATION / f"{args.model}.ckpt"
VOCODER_LOC = LOCATION / f"{args.vocoder}"
LOGO_URL = "https://shivammehta25.github.io/Matcha-TTS/images/logo.png"
assert_model_downloaded(MATCHA_TTS_LOC, MATCHA_URLS[args.model])
assert_model_downloaded(VOCODER_LOC, VOCODER_URL[args.vocoder])
device = get_device(args)

model = load_matcha(args.model, MATCHA_TTS_LOC, device)
vocoder, denoiser = load_vocoder(args.vocoder, VOCODER_LOC, device)


@torch.inference_mode()
def process_text_gradio(text):
    output = process_text(1, text, device)
    return output["x_phones"][1::2], output["x"], output["x_lengths"]


@torch.inference_mode()
def synthesise_mel(text, text_length, n_timesteps, temperature, length_scale):
    output = model.synthesise(
        text,
        text_length,
        n_timesteps=n_timesteps,
        temperature=temperature,
        spks=args.spk,
        length_scale=length_scale,
    )
    output["waveform"] = to_waveform(output["mel"], vocoder, denoiser)
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
        sf.write(fp.name, output["waveform"], 22050, "PCM_24")

    return fp.name, plot_tensor(output["mel"].squeeze().cpu().numpy())


def run_full_synthesis(text, n_timesteps, mel_temp, length_scale):
    phones, text, text_lengths = process_text_gradio(text)
    audio, mel_spectrogram = synthesise_mel(text, text_lengths, n_timesteps, mel_temp, length_scale)
    return phones, audio, mel_spectrogram


def main():
    description = """# 🍵 Matcha-TTS: A fast TTS architecture with conditional flow matching
    ### [Shivam Mehta](https://www.kth.se/profile/smehta), [Ruibo Tu](https://www.kth.se/profile/ruibo), [Jonas Beskow](https://www.kth.se/profile/beskow), [Éva Székely](https://www.kth.se/profile/szekely), and [Gustav Eje Henter](https://people.kth.se/~ghe/)
    We propose 🍵 Matcha-TTS, a new approach to non-autoregressive neural TTS, that uses conditional flow matching (similar to rectified flows) to speed up ODE-based speech synthesis. Our method:


    * Is probabilistic
    * Has compact memory footprint
    * Sounds highly natural
    * Is very fast to synthesise from


    Check out our [demo page](https://shivammehta25.github.io/Matcha-TTS). Read our [arXiv preprint for more details](https://arxiv.org/abs/2309.03199).
    Code is available in our [GitHub repository](https://github.com/shivammehta25/Matcha-TTS), along with pre-trained models.

    Cached examples are available at the bottom of the page.
    """

    with gr.Blocks(title="🍵 Matcha-TTS: A fast TTS architecture with conditional flow matching") as demo:
        processed_text = gr.State(value=None)
        processed_text_len = gr.State(value=None)

        with gr.Box():
            with gr.Row():
                gr.Markdown(description, scale=3)
                gr.Image(LOGO_URL, label="Matcha-TTS logo", height=150, width=150, scale=1, show_label=False)

        with gr.Box():
            with gr.Row():
                gr.Markdown("# Text Input")
            with gr.Row():
                text = gr.Textbox(value="", lines=2, label="Text to synthesise")

            with gr.Row():
                gr.Markdown("### Hyper parameters")
            with gr.Row():
                n_timesteps = gr.Slider(
                    label="Number of ODE steps",
                    minimum=0,
                    maximum=100,
                    step=1,
                    value=10,
                    interactive=True,
                )
                length_scale = gr.Slider(
                    label="Length scale (Speaking rate)",
                    minimum=0.5,
                    maximum=1.5,
                    step=0.05,
                    value=1.0,
                    interactive=True,
                )
                mel_temp = gr.Slider(
                    label="Sampling temperature",
                    minimum=0.00,
                    maximum=2.001,
                    step=0.16675,
                    value=0.667,
                    interactive=True,
                )

                synth_btn = gr.Button("Synthesise")

        with gr.Box():
            with gr.Row():
                gr.Markdown("### Phonetised text")
                phonetised_text = gr.Textbox(interactive=False, scale=10, label="Phonetised text")

        with gr.Box():
            with gr.Row():
                mel_spectrogram = gr.Image(interactive=False, label="mel spectrogram")

                # with gr.Row():
                audio = gr.Audio(interactive=False, label="Audio")

        with gr.Row():
            examples = gr.Examples(  # pylint: disable=unused-variable
                examples=[
                    [
                        "We propose Matcha-TTS, a new approach to non-autoregressive neural TTS, that uses conditional flow matching (similar to rectified flows) to speed up O D E-based speech synthesis.",
                        50,
                        0.677,
                        1.0,
                    ],
                    [
                        "The Secret Service believed that it was very doubtful that any President would ride regularly in a vehicle with a fixed top, even though transparent.",
                        2,
                        0.677,
                        1.0,
                    ],
                    [
                        "The Secret Service believed that it was very doubtful that any President would ride regularly in a vehicle with a fixed top, even though transparent.",
                        4,
                        0.677,
                        1.0,
                    ],
                    [
                        "The Secret Service believed that it was very doubtful that any President would ride regularly in a vehicle with a fixed top, even though transparent.",
                        10,
                        0.677,
                        1.0,
                    ],
                    [
                        "The Secret Service believed that it was very doubtful that any President would ride regularly in a vehicle with a fixed top, even though transparent.",
                        50,
                        0.677,
                        1.0,
                    ],
                    [
                        "The narrative of these events is based largely on the recollections of the participants.",
                        10,
                        0.677,
                        1.0,
                    ],
                    [
                        "The jury did not believe him, and the verdict was for the defendants.",
                        10,
                        0.677,
                        1.0,
                    ],
                ],
                fn=run_full_synthesis,
                inputs=[text, n_timesteps, mel_temp, length_scale],
                outputs=[phonetised_text, audio, mel_spectrogram],
                cache_examples=True,
            )

        synth_btn.click(
            fn=process_text_gradio,
            inputs=[
                text,
            ],
            outputs=[phonetised_text, processed_text, processed_text_len],
            api_name="matcha_tts",
            queue=True,
        ).then(
            fn=synthesise_mel,
            inputs=[processed_text, processed_text_len, n_timesteps, mel_temp, length_scale],
            outputs=[audio, mel_spectrogram],
        )

        demo.queue(concurrency_count=5).launch()


if __name__ == "__main__":
    main()