File size: 5,139 Bytes
9fd672f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import sys
import torch
import numpy as np
import traceback
from scipy.io import wavfile
import librosa
from pathlib import Path
from time import time as ttime
import shutil
from tools.my_utils import load_audio, clean_path
from feature_extractor import cnhubert

def my_save(fea, path, i_part):
    """Fix issue: torch.save doesn't support chinese path"""
    dir = os.path.dirname(path)
    name = os.path.basename(path)
    tmp_path = f"{ttime()}{i_part}.pth"
    torch.save(fea, tmp_path)
    shutil.move(tmp_path, f"{dir}/{name}")

def extract_hubert_features(data_dir="data8", exp_dir="logs/s2"):
    """Extract Hubert features for stage 2 training"""
    
    # Get project root directory (parent of GPT_SoVITS)
    root_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
    
    # Convert relative paths to absolute
    data_dir = os.path.join(root_dir, data_dir)
    exp_dir = os.path.join(root_dir, exp_dir)
    
    # Set environment variables for Hubert extraction
    inp_text = os.path.join(exp_dir, "2-name2text.txt")
    inp_wav_dir = os.path.join(exp_dir, "5-wav32k")
    exp_name = "s2"
    i_part = "0"
    all_parts = "1"
    opt_dir = exp_dir
    cnhubert.cnhubert_base_path = os.path.join(root_dir, "pretrained_models", "chinese-hubert-base")
    is_half = torch.cuda.is_available()
    
    print("Starting Hubert feature extraction...")
    print(f"Input text file: {inp_text}")
    print(f"Input wav directory: {inp_wav_dir}")
    print(f"Output directory: {opt_dir}")
    
    hubert_dir = f"{opt_dir}/4-cnhubert"
    wav32dir = f"{opt_dir}/5-wav32k"
    os.makedirs(opt_dir, exist_ok=True)
    os.makedirs(hubert_dir, exist_ok=True)
    os.makedirs(wav32dir, exist_ok=True)

    maxx = 0.95
    alpha = 0.5
    if torch.cuda.is_available():
        device = "cuda:0"
    else:
        device = "cpu"
        
    print(f"Loading Hubert model from: {cnhubert.cnhubert_base_path}")
    model = cnhubert.get_model()
    if is_half:
        model = model.half().to(device)
    else:
        model = model.to(device)

    nan_fails = []
    
    def name2go(wav_name, wav_path):
        print(f"Processing: {wav_name} from {wav_path}")
        hubert_path = f"{hubert_dir}/{wav_name}.pt"
        if os.path.exists(hubert_path):
            print(f"Skipping {wav_name} - already processed")
            return
            
        if not os.path.exists(wav_path):
            print(f"Error: WAV file not found: {wav_path}")
            return
            
        tmp_audio = load_audio(wav_path, 32000)
        if tmp_audio is None:
            print(f"Error: Failed to load audio: {wav_path}")
            return
            
        tmp_max = np.abs(tmp_audio).max()
        if tmp_max > 2.2:
            print(f"{wav_name}-filtered,{tmp_max}")
            return
            
        tmp_audio32 = (tmp_audio / tmp_max * (maxx * alpha * 32768)) + ((1 - alpha) * 32768) * tmp_audio
        tmp_audio32b = (tmp_audio / tmp_max * (maxx * alpha * 1145.14)) + ((1 - alpha) * 1145.14) * tmp_audio
        tmp_audio = librosa.resample(tmp_audio32b, orig_sr=32000, target_sr=16000)
        
        tensor_wav16 = torch.from_numpy(tmp_audio)
        if is_half:
            tensor_wav16 = tensor_wav16.half().to(device)
        else:
            tensor_wav16 = tensor_wav16.to(device)
            
        ssl = model.model(tensor_wav16.unsqueeze(0))["last_hidden_state"].transpose(1, 2).cpu()
        
        if np.isnan(ssl.detach().numpy()).sum() != 0:
            nan_fails.append((wav_name, wav_path))
            print(f"nan filtered:{wav_name}")
            return
            
        wavfile.write(
            f"{wav32dir}/{wav_name}",
            32000,
            tmp_audio32.astype("int16"),
        )
        my_save(ssl, hubert_path, i_part)
        print(f"Successfully processed {wav_name}")

    print(f"Reading text file: {inp_text}")
    with open(inp_text, "r", encoding="utf8") as f:
        lines = f.read().strip("\n").split("\n")
    print(f"Found {len(lines)} lines in text file")

    for line in lines[int(i_part)::int(all_parts)]:
        try:
            print(f"Processing line: {line}")
            wav_name, text, _, _ = line.split("\t")
            wav_name = clean_path(wav_name)
            if inp_wav_dir:
                wav_name = os.path.basename(wav_name)
                wav_path = f"{inp_wav_dir}/{wav_name}"
            else:
                wav_path = wav_name
                wav_name = os.path.basename(wav_name)
            name2go(wav_name, wav_path)
        except Exception as e:
            print(f"Error processing line: {line}")
            print(traceback.format_exc())

    if len(nan_fails) > 0 and is_half:
        print("Retrying failed files in float32 mode...")
        is_half = False
        model = model.float()
        for wav in nan_fails:
            try:
                name2go(wav[0], wav[1])
            except:
                print(f"Error retrying {wav_name}")
                print(traceback.format_exc())
    
    print("Hubert feature extraction complete.")

if __name__ == "__main__":
    extract_hubert_features()