File size: 6,230 Bytes
3a478bf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os, sys
import librosa
import soundfile as sf
import numpy as np
import re
import unicodedata
import wget
import subprocess
from pydub import AudioSegment
import tempfile
from torch import nn

import logging
from transformers import HubertModel
import warnings

# Remove this to see warnings about transformers models
warnings.filterwarnings("ignore")

logging.getLogger("fairseq").setLevel(logging.ERROR)
logging.getLogger("faiss.loader").setLevel(logging.ERROR)
logging.getLogger("transformers").setLevel(logging.ERROR)
logging.getLogger("torch").setLevel(logging.ERROR)

now_dir = os.getcwd()
sys.path.append(now_dir)

base_path = os.path.join(now_dir, "rvc", "models", "formant", "stftpitchshift")
stft = base_path + ".exe" if sys.platform == "win32" else base_path


class HubertModelWithFinalProj(HubertModel):
    def __init__(self, config):
        super().__init__(config)
        self.final_proj = nn.Linear(config.hidden_size, config.classifier_proj_size)


def load_audio(file, sample_rate):
    try:
        file = file.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
        audio, sr = sf.read(file)
        if len(audio.shape) > 1:
            audio = librosa.to_mono(audio.T)
        if sr != sample_rate:
            audio = librosa.resample(audio, orig_sr=sr, target_sr=sample_rate)
    except Exception as error:
        raise RuntimeError(f"An error occurred loading the audio: {error}")

    return audio.flatten()


def load_audio_infer(
    file, sample_rate, formant_shifting, formant_qfrency, formant_timbre
):
    try:
        file = file.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
        if not os.path.isfile(file):
            raise FileNotFoundError(f"File not found: {file}")
        audio, sr = sf.read(file)
        if len(audio.shape) > 1:
            audio = librosa.to_mono(audio.T)
        if sr != sample_rate:
            audio = librosa.resample(audio, orig_sr=sr, target_sr=sample_rate)
        if formant_shifting:
            audio = (audio * 32767).astype(np.int16)
            audio_segment = AudioSegment(
                audio.tobytes(),
                frame_rate=sample_rate,
                sample_width=2,
                channels=1,
            )
            with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_file:
                temp_file_path = temp_file.name
                audio_segment.export(temp_file_path, format="wav")

            command = [
                stft,
                "-i",
                temp_file_path,
                "-q",
                str(formant_qfrency),
                "-t",
                str(formant_timbre),
                "-o",
                f"{temp_file_path}_formatted.wav",
            ]
            subprocess.run(command, shell=True)
            formatted_audio_path = f"{temp_file_path}_formatted.wav"
            audio, sr = sf.read(formatted_audio_path)
            if len(audio.shape) > 1:
                audio = librosa.to_mono(audio.T)
            if sr != sample_rate:
                audio = librosa.resample(audio, orig_sr=sr, target_sr=sample_rate)
    except Exception as error:
        raise RuntimeError(f"An error occurred loading the audio: {error}")
    return audio.flatten()


def format_title(title):
    formatted_title = (
        unicodedata.normalize("NFKD", title).encode("ascii", "ignore").decode("utf-8")
    )
    formatted_title = re.sub(r"[\u2500-\u257F]+", "", formatted_title)
    formatted_title = re.sub(r"[^\w\s.-]", "", formatted_title)
    formatted_title = re.sub(r"\s+", "_", formatted_title)
    return formatted_title


def load_embedding(embedder_model, custom_embedder=None):
    embedder_root = os.path.join(now_dir, "rvc", "models", "embedders")
    embedding_list = {
        "contentvec": os.path.join(embedder_root, "contentvec"),
        "chinese-hubert-base": os.path.join(embedder_root, "chinese_hubert_base"),
        "japanese-hubert-base": os.path.join(embedder_root, "japanese_hubert_base"),
        "korean-hubert-base": os.path.join(embedder_root, "korean_hubert_base"),
    }

    online_embedders = {
        "contentvec": "https://huggingface.co./IAHispano/Applio/resolve/main/Resources/embedders/contentvec/pytorch_model.bin",
        "chinese-hubert-base": "https://huggingface.co./IAHispano/Applio/resolve/main/Resources/embedders/chinese_hubert_base/pytorch_model.bin",
        "japanese-hubert-base": "https://huggingface.co./IAHispano/Applio/resolve/main/Resources/embedders/japanese_hubert_base/pytorch_model.bin",
        "korean-hubert-base": "https://huggingface.co./IAHispano/Applio/resolve/main/Resources/embedders/korean_hubert_base/pytorch_model.bin",
    }

    config_files = {
        "contentvec": "https://huggingface.co./IAHispano/Applio/resolve/main/Resources/embedders/contentvec/config.json",
        "chinese-hubert-base": "https://huggingface.co./IAHispano/Applio/resolve/main/Resources/embedders/chinese_hubert_base/config.json",
        "japanese-hubert-base": "https://huggingface.co./IAHispano/Applio/resolve/main/Resources/embedders/japanese_hubert_base/config.json",
        "korean-hubert-base": "https://huggingface.co./IAHispano/Applio/resolve/main/Resources/embedders/korean_hubert_base/config.json",
    }

    if embedder_model == "custom":
        if os.path.exists(custom_embedder):
            model_path = custom_embedder
        else:
            print(f"Custom embedder not found: {custom_embedder}, using contentvec")
            model_path = embedding_list["contentvec"]
    else:
        model_path = embedding_list[embedder_model]
        bin_file = os.path.join(model_path, "pytorch_model.bin")
        json_file = os.path.join(model_path, "config.json")
        os.makedirs(model_path, exist_ok=True)
        if not os.path.exists(bin_file):
            url = online_embedders[embedder_model]
            print(f"Downloading {url} to {model_path}...")
            wget.download(url, out=bin_file)
        if not os.path.exists(json_file):
            url = config_files[embedder_model]
            print(f"Downloading {url} to {model_path}...")
            wget.download(url, out=json_file)

    models = HubertModelWithFinalProj.from_pretrained(model_path)
    return models