Hugging Face's logo

Spaces:
KdaiP
/
StableTTS1.1
Runtime error

App Files Community
StableTTS1.1
File size: 5,231 Bytes 
3dd84f8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
import os
os.environ['TMPDIR'] = './temps' # avoid the system default temp folder not having access permissions
# os.environ['HF_ENDPOINT'] = 'https://hf-mirror.com' # use huggingfacae mirror for users that could not login to huggingface

import re
import numpy as np
import matplotlib.pyplot as plt

import torch
import gradio as gr

from api import StableTTSAPI

device = 'cuda' if torch.cuda.is_available() else 'cpu'

tts_model_path = './checkpoints/checkpoint_0.pt'
vocoder_model_path = './vocoders/pretrained/firefly-gan-base-generator.ckpt'
vocoder_type = 'ffgan'

model = StableTTSAPI(tts_model_path, vocoder_model_path, vocoder_type).to(device)

@ torch.inference_mode()
def inference(text, ref_audio, language, step, temperature, length_scale, solver, cfg):
    text = remove_newlines_after_punctuation(text)
    
    if language == 'chinese':
        text = text.replace(' ', '')
        
    audio, mel = model.inference(text, ref_audio, language, step, temperature, length_scale, solver, cfg)
    
    max_val = torch.max(torch.abs(audio))
    if max_val > 1:
        audio = audio / max_val
    
    audio_output = (model.mel_config.sample_rate, (audio.cpu().squeeze(0).numpy() * 32767).astype(np.int16)) # (samplerate, int16 audio) for gr.Audio
    mel_output = plot_mel_spectrogram(mel.cpu().squeeze(0).numpy()) # get the plot of mel
    
    return audio_output, mel_output

def plot_mel_spectrogram(mel_spectrogram):
    plt.close() # prevent memory leak
    fig, ax = plt.subplots(figsize=(20, 8))
    ax.imshow(mel_spectrogram, aspect='auto', origin='lower')
    plt.axis('off')
    fig.subplots_adjust(left=0, right=1, top=1, bottom=0) # remove white edges
    return fig

def remove_newlines_after_punctuation(text):
    pattern = r'([，。！？、“”‘’《》【】；：,.!?\'\"<>()\[\]{}])\n'
    return re.sub(pattern, r'\1', text)

def main():
    
    from pathlib import Path
    examples = list(Path('./audios').rglob('*.wav'))

    # gradio wabui, reference: https://huggingface.co/spaces/fishaudio/fish-speech-1
    gui_title = 'StableTTS'
    gui_description = """Next-generation TTS model using flow-matching and DiT, inspired by Stable Diffusion 3."""
    example_text = """你指尖跳动的电光，是我永恒不变的信仰。唯我超电磁炮永世长存！"""
    
    with gr.Blocks(theme=gr.themes.Base()) as demo:
        demo.load(None, None, js="() => {const params = new URLSearchParams(window.location.search);if (!params.has('__theme')) {params.set('__theme', 'light');window.location.search = params.toString();}}")

        with gr.Row():
            with gr.Column():
                gr.Markdown(f"# {gui_title}")
                gr.Markdown(gui_description)

        with gr.Row():
            with gr.Column():
                input_text_gr = gr.Textbox(
                    label="Input Text",
                    info="Put your text here",
                    value=example_text,
                )
             
                ref_audio_gr = gr.Audio(
                    label="Reference Audio",
                    type="filepath"
                )
                
                language_gr = gr.Dropdown(
                    label='Language',
                    choices=list(model.supported_languages),
                    value = 'chinese'
                )
                
                step_gr = gr.Slider(
                    label='Step',
                    minimum=1,
                    maximum=100,
                    value=25,
                    step=1
                )
                
                temperature_gr = gr.Slider(
                    label='Temperature',
                    minimum=0,
                    maximum=2,
                    value=1,
                )
                
                length_scale_gr = gr.Slider(
                    label='Length_Scale',
                    minimum=0,
                    maximum=5,
                    value=1,
                )
                
                solver_gr = gr.Dropdown(
                    label='ODE Solver',
                    choices=['euler', 'midpoint', 'dopri5', 'rk4', 'implicit_adams', 'bosh3', 'fehlberg2', 'adaptive_heun'],
                    value = 'dopri5'
                )
                
                cfg_gr = gr.Slider(
                    label='CFG',
                    minimum=0,
                    maximum=10,
                    value=3,
                )
                
            with gr.Column():
                mel_gr = gr.Plot(label="Mel Visual")
                audio_gr = gr.Audio(label="Synthesised Audio", autoplay=True)
                tts_button = gr.Button("\U0001F3A7 Generate / 合成", elem_id="send-btn", visible=True, variant="primary")
                examples = gr.Examples(examples, ref_audio_gr)

        tts_button.click(inference, [input_text_gr, ref_audio_gr, language_gr, step_gr, temperature_gr, length_scale_gr, solver_gr, cfg_gr], outputs=[audio_gr, mel_gr])

    demo.queue()  
    demo.launch(debug=True, show_api=True)


if __name__ == '__main__':
    main()