File size: 10,235 Bytes
1cb796d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b4f6525
 
1cb796d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6fe5063
db06f79
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1cb796d
 
 
 
edf08a9
50600ce
2f97a56
deb1df1
1cb796d
 
3444577
1cb796d
 
 
 
 
db06f79
1cb796d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
50600ce
 
94a07c5
50600ce
db06f79
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
212
213
214
import os
import glob
import json
import argparse
import traceback
import logging
import gradio as gr
import numpy as np
import librosa
import torch
import asyncio
import edge_tts
from datetime import datetime
from fairseq import checkpoint_utils
from infer_pack.models import SynthesizerTrnMs256NSFsid, SynthesizerTrnMs256NSFsid_nono
from vc_infer_pipeline import VC
from config import Config
config = Config()
logging.getLogger("numba").setLevel(logging.WARNING)
limitation = os.getenv("SYSTEM") == "spaces"  # limit audio length in huggingface spaces

def create_vc_fn(tgt_sr, net_g, vc, if_f0, file_index):
    def vc_fn(
        input_audio,
        f0_up_key,
        f0_method,
        index_rate,
        tts_mode,
        tts_text,
        tts_voice
    ):
        try:
            if tts_mode:
                if len(tts_text) > 100 and limitation:
                    return "Text is too long", None
                if tts_text is None or tts_voice is None:
                    return "You need to enter text and select a voice", None
                asyncio.run(edge_tts.Communicate(tts_text, "-".join(tts_voice.split('-')[:-1])).save("tts.mp3"))
                audio, sr = librosa.load("tts.mp3", sr=16000, mono=True)
            else:
                if config.files:
                    audio, sr = librosa.load(input_audio, sr=16000, mono=True)
                else:
                    if input_audio is None:
                        return "You need to upload an audio", None
                    sampling_rate, audio = input_audio
                    duration = audio.shape[0] / sampling_rate
                    if duration > 20 and limitation:
                        return "Please upload an audio file that is less than 20 seconds. If you need to generate a longer audio file, please use Colab.", None
                    audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
                    if len(audio.shape) > 1:
                        audio = librosa.to_mono(audio.transpose(1, 0))
                    if sampling_rate != 16000:
                        audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
            times = [0, 0, 0]
            f0_up_key = int(f0_up_key)
            audio_opt = vc.pipeline(
                hubert_model,
                net_g,
                0,
                audio,
                times,
                f0_up_key,
                f0_method,
                file_index,
                index_rate,
                if_f0,
                f0_file=None,
            )
            print(
                f"[{datetime.now().strftime('%Y-%m-%d %H:%M')}]: npy: {times[0]}, f0: {times[1]}s, infer: {times[2]}s"
            )
            return "Success", (tgt_sr, audio_opt)
        except:
            info = traceback.format_exc()
            print(info)
            return info, (None, None)
    return vc_fn

def load_hubert():
    global hubert_model
    models, _, _ = checkpoint_utils.load_model_ensemble_and_task(
        ["hubert_base.pt"],
        suffix="",
    )
    hubert_model = models[0]
    hubert_model = hubert_model.to(config.device)
    if config.is_half:
        hubert_model = hubert_model.half()
    else:
        hubert_model = hubert_model.float()
    hubert_model.eval()

def change_to_tts_mode(tts_mode):
    if tts_mode:
        return gr.Audio.update(visible=False), gr.Textbox.update(visible=True), gr.Dropdown.update(visible=True)
    else:
        return gr.Audio.update(visible=True), gr.Textbox.update(visible=False), gr.Dropdown.update(visible=False)

if __name__ == '__main__':
    load_hubert()
    models = []
    tts_voice_list = asyncio.get_event_loop().run_until_complete(edge_tts.list_voices())
    voices = [f"{v['ShortName']}-{v['Gender']}" for v in tts_voice_list]
    if limitation:
        with open("weights/model_info.json", "r", encoding="utf-8") as f:
            models_info = json.load(f)
        for name, info in models_info.items():
            if not info['enable']:
                continue
            title = info['title']
            author = info.get("author", None)
            cover = f"weights/{name}/{info['cover']}"
            index = f"weights/{name}/{info['feature_retrieval_library']}"
            cpt = torch.load(f"weights/{name}/{name}.pth", map_location="cpu")
            tgt_sr = cpt["config"][-1]
            cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0]  # n_spk
            if_f0 = cpt.get("f0", 1)
            if if_f0 == 1:
                net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half)
            else:
                net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
            del net_g.enc_q
            print(net_g.load_state_dict(cpt["weight"], strict=False))  # 不加这一行清不干净, 真奇葩
            net_g.eval().to(config.device)
            if config.is_half:
                net_g = net_g.half()
            else:
                net_g = net_g.float()
            vc = VC(tgt_sr, config)
            models.append((name, title, author, cover, create_vc_fn(tgt_sr, net_g, vc, if_f0, index)))
    else:
        folder_path = "weights"
        for name in os.listdir(folder_path):
            print("check folder: " + name)
            if name.startswith("."): break
            cover_path = glob.glob(f"{folder_path}/{name}/*.png") + glob.glob(f"{folder_path}/{name}/*.jpg")
            index_path = glob.glob(f"{folder_path}/{name}/*.index")
            checkpoint_path = glob.glob(f"{folder_path}/{name}/*.pth")
            title = name
            author = ""
            if cover_path:
                cover = cover_path[0]
            else:
                cover = ""
            index = index_path[0]
            cpt = torch.load(checkpoint_path[0], map_location="cpu")
            tgt_sr = cpt["config"][-1]
            cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0]  # n_spk
            if_f0 = cpt.get("f0", 1)
            if if_f0 == 1:
                net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half)
            else:
                net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
            del net_g.enc_q
            print(net_g.load_state_dict(cpt["weight"], strict=False))  # 不加这一行清不干净, 真奇葩
            net_g.eval().to(config.device)
            if config.is_half:
                net_g = net_g.half()
            else:
                net_g = net_g.float()
            vc = VC(tgt_sr, config)
            models.append((name, title, author, cover, create_vc_fn(tgt_sr, net_g, vc, if_f0, index)))
    with gr.Blocks() as app:
        gr.Markdown(
            "# <center> RVC Models (Latest Update)\n"
            "## <center> The input audio should be clean and pure voice without background music.\n"
            "### <center> [Recommended to use google colab for more features](https://colab.research.google.com/drive/110kiMZTdP6Ri1lY9-NbQf17GVPPhHyeT?usp=sharing) \n"
            "#### <center> Please regenerate your model to latest RVC to fully applied this new rvc.\n"
            "[![image](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/110kiMZTdP6Ri1lY9-NbQf17GVPPhHyeT?usp=sharing)\n\n"
            "[![Original Repo](https://badgen.net/badge/icon/github?icon=github&label=Original%20Repo)](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI)"
        )
        with gr.Tabs():
            for (name, title, author, cover, vc_fn) in models:
                with gr.TabItem(name):
                    with gr.Row():
                        gr.Markdown(
                            '<div align="center">'
                            f'<div>{title}</div>\n'+
                            (f'<div>Model author: {author}</div>' if author else "")+
                            (f'<img style="width:auto;height:300px;" src="file/{cover}">' if cover else "")+
                            '</div>'
                        )
                    with gr.Row():
                        with gr.Column():
                            vc_input = gr.Audio(label="Input audio"+' (less than 20 seconds)' if limitation else '')
                            vc_transpose = gr.Number(label="Transpose", value=0)
                            vc_f0method = gr.Radio(
                                label="Pitch extraction algorithm, PM is fast but Harvest is better for low frequencies",
                                choices=["pm", "harvest"],
                                value="pm",
                                interactive=True,
                            )
                            vc_index_ratio = gr.Slider(
                                minimum=0,
                                maximum=1,
                                label="Retrieval feature ratio",
                                value=0.6,
                                interactive=True,
                            )
                            tts_mode = gr.Checkbox(label="tts (use edge-tts as input)", value=False)
                            tts_text = gr.Textbox(visible=False,label="TTS text (100 words limitation)" if limitation else "TTS text")
                            tts_voice = gr.Dropdown(label="Edge-tts speaker", choices=voices, visible=False, allow_custom_value=False, value="en-US-AnaNeural-Female")
                            vc_submit = gr.Button("Generate", variant="primary")
                        with gr.Column():
                            vc_output1 = gr.Textbox(label="Output Message")
                            vc_output2 = gr.Audio(label="Output Audio")
                vc_submit.click(vc_fn, [vc_input, vc_transpose, vc_f0method, vc_index_ratio, tts_mode, tts_text, tts_voice], [vc_output1, vc_output2])
                tts_mode.change(change_to_tts_mode, [tts_mode], [vc_input, tts_text, tts_voice])
        gr.Markdown('# <center>Changelog 2023.05.15')
        gr.Markdown('- Added support for direct upload to gradio')
        gr.Markdown('- Added ayato-jp and eula-jp')
        gr.Markdown('- Minor fix and adjustment')
        app.queue(concurrency_count=1, max_size=20, api_open=config.api).launch(share=config.colab)