-
Notifications
You must be signed in to change notification settings - Fork 16
/
Copy pathtts_audio_normalizer.py
310 lines (261 loc) · 11.2 KB
/
tts_audio_normalizer.py
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
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
import os
import logging
from concurrent.futures import ThreadPoolExecutor, as_completed
from datetime import datetime
from typing import Any, Dict, List, Optional, Tuple
from dataclasses import dataclass
import numpy as np
import librosa
import soundfile as sf
import sox
import noisereduce as nr
from tqdm import tqdm
@dataclass
class AudioProcessingParams:
def __init__(self):
# 基础参数
self.rate = 44100
self.channels = 1
self.rate_enabled = True
self.channels_enabled = True
# 降噪参数
self.noise_reduction_enabled = True
self.noise_reduction_strength = 0.7
self.noise_attack_time = 0.02
self.noise_release_time = 0.02
# 静音处理参数
self.silence_threshold = 0.1
self.min_silence_duration = 0.3
# 频率响应参数
self.subsonic_filter_enabled = True
self.subsonic_frequency = 60
self.equalizer_enabled = True
self.bass_gain = 0
self.bass_frequency = 100
self.bass_slope = 0.5
self.mid_frequency = 1000
self.mid_q = 0.707
self.mid_gain = 0
self.treble_gain = 0
self.treble_frequency = 5000
self.treble_slope = 0.5
# 动态范围参数
self.compand_enabled = True
self.attack_time = 0.02
self.decay_time = 0.15
self.soft_knee_db = 6.0
self.target_db = -3
# 淡入淡出参数
self.fade_enabled = True
self.fade_in_time = 0.01
self.fade_out_time = 0.01
# 输出格式
self.output_format = "wav"
class TTSAudioNormalizer:
"""专门针对TTS训练的音频标准化处理器"""
def __init__(self):
self.quality_metrics: Dict[str, Any] = {}
self.supported_formats = ['.wav', '.mp3', '.flac', '.ogg']
self.processed_count = 0
self.error_count = 0
def normalize_audio(self,
input_path: str,
output_path: str,
params: Optional[AudioProcessingParams] = None) -> bool:
"""
优化的音频处理流程，专门针对TTS训练
"""
if params is None:
params = AudioProcessingParams()
try:
self._validate_input_file(input_path)
os.makedirs(os.path.dirname(output_path), exist_ok=True)
# 1. 首先进行降噪处理
temp_path = None
if params.noise_reduction_enabled:
# 读取音频文件
y, sr = librosa.load(input_path, sr=None)
# 使用noisereduce进行降噪
reduced_noise = nr.reduce_noise(
y=y,
sr=sr,
prop_decrease=params.noise_reduction_strength,
stationary=True,
n_jobs=-1
)
# 保存临时文件
temp_path = output_path + '.temp.wav'
sf.write(temp_path, reduced_noise, sr)
# 更新input_path为降噪后的临时文件
processing_input = temp_path
else:
processing_input = input_path
# 2. 使用sox进行其他处理
transformer = sox.Transformer()
# 基础预处理
if params.rate_enabled:
transformer.rate(params.rate)
if params.channels_enabled:
transformer.channels(params.channels)
transformer.dcshift(shift=0.0)
# 动态范围压缩
if params.compand_enabled:
transformer.compand(
attack_time=params.attack_time,
decay_time=params.decay_time,
soft_knee_db=params.soft_knee_db
)
# 静音处理
transformer.silence( # 去除开头的静音
location=1,
silence_threshold=params.silence_threshold,
min_silence_duration=params.min_silence_duration,
buffer_around_silence=True
)
transformer.silence( # 去除结尾的静音
location=-1,
silence_threshold=params.silence_threshold,
min_silence_duration=params.min_silence_duration,
buffer_around_silence=True
)
transformer.silence( # 去除中间的静音段
location=0,
silence_threshold=params.silence_threshold,
min_silence_duration=params.min_silence_duration,
buffer_around_silence=True
)
# 频率响应优化
if params.subsonic_filter_enabled:
transformer.highpass(params.subsonic_frequency, 2.0)
if params.equalizer_enabled:
if params.bass_gain != 0:
transformer.bass(
gain_db=params.bass_gain,
frequency=params.bass_frequency,
slope=params.bass_slope
)
transformer.equalizer(
frequency=params.mid_frequency,
width_q=params.mid_q,
gain_db=params.mid_gain
)
if params.treble_gain != 0:
transformer.treble(
gain_db=params.treble_gain,
frequency=params.treble_frequency,
slope=params.treble_slope
)
# 音量标准化
transformer.norm(params.target_db)
# 淡入淡出
if params.fade_enabled:
transformer.fade(
fade_in_len=params.fade_in_time,
fade_out_len=params.fade_out_time
)
# 执行转换
transformer.build(processing_input, output_path)
# 删除临时文件
if temp_path and os.path.exists(temp_path):
os.remove(temp_path)
# 质量检查
self._check_audio_quality(output_path)
self.processed_count += 1
logging.info(f"成功处理文件: {input_path}")
return True
except Exception as e:
self._handle_error(input_path, e)
self.error_count += 1
return False
def _validate_input_file(self, input_path: str) -> None:
"""验证输入文件"""
if not os.path.exists(input_path):
raise FileNotFoundError(f"输入文件不存在: {input_path}")
if not any(input_path.lower().endswith(fmt) for fmt in self.supported_formats):
raise ValueError(f"不支持的文件格式: {input_path}")
def _handle_error(self, input_path: str, error: Exception) -> None:
"""处理错误"""
logging.error(f"处理文件失败 {input_path}: {str(error)}")
def _check_audio_quality(self, audio_path: str) -> None:
"""检查处理后的音频质量"""
try:
y, sr = librosa.load(audio_path, sr=None)
metrics = {
'duration': librosa.get_duration(y=y, sr=sr),
'rms_energy': float(np.sqrt(np.mean(y**2))),
'zero_crossings': int(librosa.zero_crossings(y).sum()),
'spectral_centroid': float(np.mean(librosa.feature.spectral_centroid(y=y, sr=sr))),
'silence_ratio': float(np.mean(np.abs(y) < 0.01))
}
self.quality_metrics[audio_path] = metrics
if metrics['duration'] < 0.1:
raise ValueError("音频太短")
if metrics['rms_energy'] < 0.01:
raise ValueError("音量太小")
if metrics['silence_ratio'] > 0.5:
raise ValueError("静音比例过高")
except Exception as e:
logging.warning(f"质量检查失败 {audio_path}: {str(e)}")
def _get_audio_files(self, input_dir: str, output_dir: str, params: AudioProcessingParams) -> List[Tuple[str, str]]:
"""获取需要处理的音频文件列表"""
audio_files = []
for root, _, files in os.walk(input_dir):
for file in files:
if any(file.lower().endswith(fmt) for fmt in self.supported_formats):
input_path = os.path.join(root, file)
rel_path = os.path.relpath(input_path, input_dir)
output_path = os.path.join(
output_dir,
os.path.splitext(rel_path)[0] + '.' + params.output_format
)
audio_files.append((input_path, output_path))
return audio_files
def batch_normalize_directory(self,
input_dir: str,
output_dir: str,
params: Optional[AudioProcessingParams] = None,
max_workers: int = 4) -> None:
"""批量处理目录中的音频文件"""
if params is None:
params = AudioProcessingParams()
if not os.path.exists(input_dir):
raise FileNotFoundError(f"输入目录不存在: {input_dir}")
audio_files = self._get_audio_files(input_dir, output_dir, params)
total_files = len(audio_files)
with ThreadPoolExecutor(max_workers=max_workers) as executor:
with tqdm(total=total_files, desc="处理进度", ncols=100) as pbar:
def process_file(args):
input_path, output_path = args
result = self.normalize_audio(input_path, output_path, params)
pbar.update(1)
return result
list(executor.map(process_file, audio_files))
self._print_summary(total_files)
self._print_quality_summary()
def _print_summary(self, total_files: int) -> None:
"""打印处理总结"""
print("\n处理完成:")
print(f"总文件数: {total_files}")
print(f"成功处理: {self.processed_count}")
print(f"处理失败: {self.error_count}")
def _print_quality_summary(self) -> None:
"""打印音频质量统计报告"""
if not self.quality_metrics:
return
print("\n音频质量统计:")
metrics_summary = {
'duration': [],
'rms_energy': [],
'silence_ratio': []
}
for metrics in self.quality_metrics.values():
metrics_summary['duration'].append(metrics['duration'])
metrics_summary['rms_energy'].append(metrics['rms_energy'])
metrics_summary['silence_ratio'].append(metrics['silence_ratio'])
for metric_name, values in metrics_summary.items():
values = np.array(values)
print(f"\n{metric_name}统计:")
print(f" 平均值: {np.mean(values):.3f}")
print(f" 标准差: {np.std(values):.3f}")
print(f" 最小值: {np.min(values):.3f}")
print(f" 最大值: {np.max(values):.3f}")