neverlib 0.2.5__py3-none-any.whl → 0.2.7__py3-none-any.whl

neverlib/QA/gen_init.py

@@ -0,0 +1,117 @@
+# 自动生成指定包目录下的 __init__.py（懒加载格式：lazy_loader.attach）
+
+import ast
+from pathlib import Path
+from typing import Dict, List, Set
+
+
+def _extract_exports_from_module(py_file: Path) -> List[str]:
+    """
+    优先读取模块内显式定义的 __all__，否则收集顶层的类与函数名（排除以下划线开头的）。
+    """
+    try:
+        source = py_file.read_text(encoding='utf-8')
+    except UnicodeDecodeError:
+        source = py_file.read_text(errors='ignore')
+
+    try:
+        tree = ast.parse(source, filename=str(py_file))
+    except SyntaxError:
+        return []
+
+    # 1) 如果定义了 __all__ 则优先使用
+    for node in tree.body:
+        if isinstance(node, ast.Assign):
+            for target in node.targets:
+                if isinstance(target, ast.Name) and target.id == '__all__':
+                    values: List[str] = []
+                    if isinstance(node.value, (ast.List, ast.Tuple)):
+                        for elt in node.value.elts:
+                            if isinstance(elt, ast.Str):  # py<3.8
+                                values.append(elt.s)
+                            elif isinstance(elt, ast.Constant) and isinstance(elt.value, str):
+                                values.append(elt.value)
+                    return [name for name in values if name and not name.startswith('_')]
+
+    # 2) 否则收集顶层函数与类名
+    exports: Set[str] = set()
+    for node in tree.body:
+        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
+            name = node.name
+            if not name.startswith('_'):
+                exports.add(name)
+    return sorted(exports)
+
+
+def generate_init_for_directory(package_dir: Path) -> Path:
+    """
+    为指定目录生成懒加载版 __init__.py（覆盖写入）。
+    - 仅扫描一级子模块（同级 .py 文件），忽略以下划线开头的模块与 __init__.py 本身。
+    - 生成结构：
+        from lazy_loader import attach
+
+        __getattr__, __dir__, __all__ = attach(
+            __name__,
+            submodules=["m1", "m2", ...],
+            submod_attrs={"m1": ["A", "B"], ...}
+        )
+    """
+    package_dir = package_dir.resolve()
+    if not package_dir.is_dir():
+        raise ValueError(f"目标不是目录: {package_dir}")
+
+    module_to_exports: Dict[str, List[str]] = {}
+    module_names: List[str] = []
+    for py_file in sorted(package_dir.glob('*.py')):
+        if py_file.name == '__init__.py':
+            continue
+        if py_file.name.startswith('_'):
+            continue
+        module_name = py_file.stem
+        module_names.append(module_name)
+        exports = _extract_exports_from_module(py_file)
+        if exports:
+            module_to_exports[module_name] = exports
+
+    lines: List[str] = []
+    lines.append('# This file is auto-generated. Do NOT edit manually.')
+    lines.append('# Generated by neverlib.QA.gen_init')
+    lines.append('from lazy_loader import attach')
+    lines.append('')
+    lines.append('__getattr__, __dir__, __all__ = attach(')
+    lines.append('    __name__,')
+    lines.append('    submodules=[')
+    for name in sorted(module_names):
+        lines.append(f'        "{name}",')
+    lines.append('    ],')
+    lines.append('    submod_attrs={')
+    for module_name in sorted(module_to_exports.keys()):
+        exports_sorted = sorted(module_to_exports[module_name])
+        joined = ', '.join([f"'{e}'" for e in exports_sorted])
+        lines.append(f'        "{module_name}": [{joined}],')
+    lines.append('    }')
+    lines.append(')')
+    lines.append('')
+
+    init_file = package_dir / '__init__.py'
+    init_file.write_text('\n'.join(lines), encoding='utf-8')
+    return init_file
+
+
+def main():
+    import argparse
+
+    parser = argparse.ArgumentParser(description='为指定包目录自动生成 __init__.py')
+    parser.add_argument('-p', '--package-dir', dest='package_dir',
+                        default='../../neverlib/vad',
+                        help='包目录路径，比如 /path/to/pkg 或 ./neverlib/utils')
+    args = parser.parse_args()
+
+    target_dir = Path(args.package_dir)
+    init_path = generate_init_for_directory(target_dir)
+    print(f'已生成: {init_path}')
+# python -m neverlib.QA.gen_init /data03/never/Desktop/neverlib/neverlib/utils
+
+
+if __name__ == '__main__':
+    main()

neverlib/QA/get_fun.py

@@ -0,0 +1,19 @@
+# 获取一个python脚本里面所有的函数名
+
+import ast
+from typing import List
+
+
+def get_function_names(file_path: str) -> List[str]:
+    with open(file_path, 'r', encoding='utf-8') as file:
+        source = file.read()
+    tree = ast.parse(source, filename=file_path)
+    names: List[str] = []
+    for node in ast.walk(tree):
+        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)):
+            names.append(node.name)
+    return names
+
+
+if __name__ == '__main__':
+    print(get_function_names('../utils/checkGPU.py'))

neverlib/__init__.py

@@ -1,15 +1,31 @@
 '''
 Author: 凌逆战 | Never
-Date: 2025-08-22
+Date: 2025-09-07
 Description: neverlib - 音频处理和VAD工具集
+
+这是一个提供音频处理、增强、分析和语音活动检测(VAD)功能的Python库。
+该库使用懒加载机制，可以根据需要导入模块，提高启动速度并减少内存占用。
+
+主要功能模块:
+- utils: 实用工具函数
+- vad: 语音活动检测
+- audio_aug: 音频增强和数据增广
+- filter: 滤波和音频处理
+- data_analyze: 数据分析工具
+- metrics: 音频质量评估指标
+
+注意: 所有功能需要通过具体子模块导入，例如:
+  from neverlib.audio_aug import limiter
+  from neverlib.vad import EnergyVad_C
+  from neverlib.filter import HPFilter
 '''
 try:
     import re
     import pathlib
-    
+
     # 获取pyproject.toml的路径
     _pyproject_path = pathlib.Path(__file__).parent.parent / "pyproject.toml"
-    
+
     # 读取版本号
     if _pyproject_path.exists():
         with open(_pyproject_path, "r", encoding="utf-8") as f:
@@ -27,5 +43,6 @@ from lazy_loader import attach
 
 __getattr__, __dir__, __all__ = attach(
     __name__,
-    submodules=["utils", "vad", "audio_aug", "filter", "data_analyze"],
+    submodules=["audio_aug", "data_analyze", "filter", "metrics", "utils", "vad", ],
+    # 只导出子模块，不直接导出函数
 )

neverlib/audio_aug/HarmonicDistortion.py

@@ -3,10 +3,8 @@ Author: 凌逆战 | Never
 Date: 2025-07-29 17:49:25
 Description: 
 '''
-import pedalboard as pdb
-import librosa
-import soundfile as sf
 import numpy as np
+import soundfile as sf
 
 
 def apply_harmonic_distortion(wav, drive=1.0, mix=1.0):
@@ -54,11 +52,15 @@ def apply_pedalboard_distortion(wav, sr, drive_db=15.0):
     sr (int): 采样率。
     drive_db (float): 驱动增益, 单位是分贝(dB)。值越大失真越严重。
     """
+    try:
+        import pedalboard as pdb
+    except ImportError:
+        raise ImportError(
+            "pedalboard is required for apply_pedalboard_distortion(). "
+            "Please install it via `pip install pedalboard`.")
     # 1. 创建一个效果器处理板 (Pedalboard)
     # 这里只放一个 Distortion 效果器
-    board = pdb.Pedalboard([
-        pdb.Distortion(drive_db=drive_db)
-    ])
+    board = pdb.Pedalboard([pdb.Distortion(drive_db=drive_db)])
 
     # 2. 处理音频
     # pedalboard 要求输入是 (channels, samples) 或 (samples,)
@@ -68,12 +70,16 @@ def apply_pedalboard_distortion(wav, sr, drive_db=15.0):
     return distorted_wav
 
 
-# --- 使用示例 ---
-y, sr = librosa.load('your_audio.wav', sr=None)
+if __name__ == "__main__":
+
+    # --- 使用示例 ---
+    y, sr = sf.read('your_audio.wav', sr=None)
 
-# 模拟一个中等程度的过载失真
-drive_db_amount = 25.0
-y_pb_distorted = apply_pedalboard_distortion(y, sr, drive_db=drive_db_amount)
+    # 模拟一个中等程度的过载失真
+    drive_db_amount = 25.0
+    y_pb_distorted = apply_pedalboard_distortion(y,
+                                                 sr,
+                                                 drive_db=drive_db_amount)
 
-sf.write('augmented_pedalboard_distortion.wav', y_pb_distorted, sr)
-print("使用 Pedalboard 的谐波失真增强完成！")
+    sf.write('augmented_pedalboard_distortion.wav', y_pb_distorted, sr)
+    print("使用 Pedalboard 的谐波失真增强完成！")

neverlib/audio_aug/__init__.py

@@ -1,12 +1,30 @@
-# -*- coding:utf-8 -*-
-# Author:凌逆战 | Never
-# Date: 2024/5/17
-"""
-音频增强模块
-"""
-from lazy_loader import attach
-
-__getattr__, __dir__, __all__ = attach(
-    __name__,
-    submodules=["audio_aug"],
-)
+# This file is auto-generated. Do NOT edit manually.
+# Generated by neverlib.QA.gen_init
+from lazy_loader import attach
+
+__getattr__, __dir__, __all__ = attach(
+    __name__,
+    submodules=[
+        "HarmonicDistortion",
+        "TFMask",
+        "audio_aug",
+        "clip_aug",
+        "coder_aug",
+        "coder_aug2",
+        "loss_packet_aug",
+        "quant_aug",
+    ],
+    submod_attrs={
+        "HarmonicDistortion": ['apply_harmonic_distortion', 'apply_pedalboard_distortion'],
+        "TFMask": ['FreqMask', 'TimeMask'],
+        "audio_aug": ['add_reverb', 'get_snr_use_vad', 'limiter', 'measure_loudness', 'snr_aug_Interpolation',
+                      'snr_aug_changeClean', 'snr_aug_changeNoise', 'snr_aug_vad_Interpolation',
+                      'snr_diff_changeClean', 'snr_diff_changeNoise', 'volume_aug', 'volume_aug_dbrms',
+                      'volume_aug_linmax', 'volume_aug_lufs', 'volume_convert', 'volume_norm'],
+        "clip_aug": ['clipping_aug'],
+        "coder_aug": ['aac_aug_save', 'amr_nb_aug', 'amr_wb_aug', 'flac_aug', 'flac_encode_save', 'mp3_aug', 'opus_aug_save', 'vorbis_aug'],
+        "coder_aug2": ['apply_codec_distortion', 'check_codec_available'],
+        "loss_packet_aug": ['simulate_packet_loss_vectorized'],
+        "quant_aug": ['apply_mulaw_quantization', 'apply_uniform_quantization'],
+    }
+)

neverlib/audio_aug/audio_aug.py

@@ -6,10 +6,9 @@
 """
 import random
 import numpy as np
-import soundfile as sf
 from scipy import signal
-from neverlib.utils import EPS
-from neverlib.filter import HPFilter
+from ..utils import EPS
+from ..filter import HPFilter
 
 
 def limiter(wav, threshold=0.999):
@@ -299,18 +298,18 @@ def volume_aug_linmax(wav, target_level, hpf=False, sr=16000, order=4, cutoff=10
     return wav_opt
 
 
-import pyloudnorm as pyln
+# 注意: 避免在模块导入阶段引入可选依赖 pyloudnorm
 
 def volume_aug_lufs(wav, target_lufs, hpf=False, sr=16000, order=4, cutoff=100):
     """
     音量增强, 使用lufs方法, 
     LUFS是“感知响度” → 跟人耳听感对齐，而且符合国际响度标准。
-    
+
     LUFS 使用 感知加权（K-weighting）
     - 高频增强（模拟人耳在 3~6kHz 的敏感）
     - 低频衰减（降低 <100Hz 对响度的影响）。
     使用 短时块（400ms）能量 + 响度门限（-70 LUFS） 过滤极静音段。
-    
+
     Args:
         wav: 音频
         target_lufs: 目标音量, 单位lufs
@@ -318,7 +317,7 @@ def volume_aug_lufs(wav, target_lufs, hpf=False, sr=16000, order=4, cutoff=100):
         sr: 采样率
         order: 滤波器阶数
         cutoff: 截止频率
-        
+
     补充信息: 
     ## 推荐的 target_lufs 值（行业参考）
     平台	推荐目标 LUFS
@@ -328,6 +327,11 @@ def volume_aug_lufs(wav, target_lufs, hpf=False, sr=16000, order=4, cutoff=100):
     游戏音频	-16 ~ -18
     有声书	-18 ~ -20
     """
+    try:
+        import pyloudnorm as pyln
+    except Exception as e:
+        raise ImportError("需要安装 pyloudnorm 才能使用 volume_aug_lufs: pip install pyloudnorm") from e
+
     wav_tmp = wav.copy()
     if hpf:
         wav_tmp = HPFilter(wav_tmp, sr=sr, order=4, cutoff=1000)
@@ -336,16 +340,17 @@ def volume_aug_lufs(wav, target_lufs, hpf=False, sr=16000, order=4, cutoff=100):
     meter = pyln.Meter(sr, block_size=0.400)  # block_size=400ms
 
     # Step3: 测量当前 LUFS
-    loudness = meter.integrated_loudness(wav_tmp)   
+    loudness = meter.integrated_loudness(wav_tmp)
 
     # Step4: 计算增益并应用
     loudness_diff = target_lufs - loudness
     scalar = 10 ** (loudness_diff / 20.0)
     wav_opt = wav * scalar
 
-    wav_opt = limiter(wav_opt, threshold=0.999) # Step5: 限幅
+    wav_opt = limiter(wav_opt, threshold=0.999)  # Step5: 限幅
     return wav_opt
 
+
 def measure_loudness(wav, sr):
     """
     测量音频的 Peak / RMS / LUFS，以及峰均比（Crest Factor）
@@ -384,10 +389,10 @@ def measure_loudness(wav, sr):
         "lufs": loudness_lufs,
         "crest_factor_db": crest_factor_db
     }
-    
-    
-def volume_convert(value, 
-                   from_unit="linear", to_unit="dBFS", 
+
+
+def volume_convert(value,
+                   from_unit="linear", to_unit="dBFS",
                    crest_factor_db=None, lufs_offset=None):
     """
     音量单位转换函数
@@ -442,4 +447,4 @@ def volume_convert(value,
         rms_dbfs = 20 * np.log10(lin_val + EPS)
         return rms_dbfs - lufs_offset
     else:
-        raise ValueError(f"未知单位：{to_unit}")
+        raise ValueError(f"未知单位：{to_unit}")

neverlib/audio_aug/clip_aug.py

@@ -3,13 +3,9 @@ Author: 凌逆战 | Never
 Date: 2025-07-29 17:06:28
 Description: 
 '''
-import sys
-sys.path.append("..")
-import librosa
+import random
 import numpy as np
 import soundfile as sf
-import random
-from audiomentations import Compose, ClippingDistortion
 
 
 def clipping_aug(wav):
@@ -17,25 +13,26 @@ def clipping_aug(wav):
     模拟录音设备或音频处理设备的动态范围限制
     """
     wav = wav / np.max(np.abs(wav))  # 归一化
-    gain = random.uniform(1.0, 2)   # 增益
+    gain = random.uniform(1.0, 2)  # 增益
     wav = wav * gain
     wav = np.clip(wav, -1.0, 1.0)
 
     return wav
 
 
-wav_path = "/data/never/Desktop/kws_train/QA/wav_data/TIMIT.wav"
-wav, wav_sr = sf.read(wav_path, always_2d=True)
-wav = wav.T
-print(wav.shape)
+if __name__ == "__main__":
+    wav_path = "/data/never/Desktop/kws_train/QA/wav_data/TIMIT.wav"
+    wav, wav_sr = sf.read(wav_path, always_2d=True)
+    wav = wav.T
+    print(wav.shape)
 
-# 应用削波增强
-# 我们让削波阈值在音频振幅的50%到75%之间随机选择
-# 这意味着信号中最响亮的25%到50%的部分将被削平
-y_clipped = clipping_aug(wav, wav_sr, min_percentile=50, max_percentile=75)
+    # 应用削波增强
+    # 我们让削波阈值在音频振幅的50%到75%之间随机选择
+    # 这意味着信号中最响亮的25%到50%的部分将被削平
+    y_clipped = clipping_aug(wav, wav_sr, min_percentile=50, max_percentile=75)
 
-# 保存增强后的音频
-output_path = './augmented_clipped.wav'
-sf.write(output_path, y_clipped.T, wav_sr)
+    # 保存增强后的音频
+    output_path = './augmented_clipped.wav'
+    sf.write(output_path, y_clipped.T, wav_sr)
 
-print(f"削波增强完成！增强后的音频已保存至: {output_path}")
+    print(f"削波增强完成！增强后的音频已保存至: {output_path}")

neverlib/audio_aug/coder_aug.py

@@ -20,16 +20,20 @@ AMR (Adaptive Multi-Rate)
 - 压缩特性：严格为语音设计, 会滤除大部分非语音频率（如音乐）, 导致音乐听起来“电话音”效果。
 - 数据增强目的：固定采样率：AMR-NB (窄带) 为 8kHz, AMR-WB (宽带) 为 16kHz。这一点至关重要！
 """
+import random
 import numpy as np
 import soundfile as sf
-from audiomentations import Mp3Compression
-import av
-import random
-
-# mp3编解码数据增强
 
 
 def mp3_aug(wav, sr):
+    # mp3编解码数据增强
+    try:
+        from audiomentations import Mp3Compression
+    except ImportError:
+        raise ImportError(
+            "audiomentations is required for mp3_aug(). "
+            "Please install it via `pip install audiomentations`.")
+
     # return Mp3Compression(min_bitrate=64, max_bitrate=192, p=1.0)(samples, sample_rate)
     return sf.write('audio.mp3', wav, sr, format='MP3', bitrate='192k')
 
@@ -47,11 +51,20 @@ def opus_aug_save(wav: np.ndarray, sr: int, output_filepath: str):
     对音频进行 Opus 压缩, 并直接保存到文件。
     使用 PyAV 实现, 比特率是随机的。
     """
+    try:
+        import av
+    except ImportError:
+        raise ImportError("av is required for opus_aug_save(). "
+                          "Please install it via `pip install av`.")
+
     # 随机选择一个比特率 (kbps)
     bitrate_kbps = random.choice([24, 32, 48, 64, 96, 128])
-    output_filepath_with_bitrate = output_filepath.replace('.opus', f'_{bitrate_kbps}k.opus')
+    output_filepath_with_bitrate = output_filepath.replace(
+        '.opus', f'_{bitrate_kbps}k.opus')
 
-    print(f"  -> Saving Opus augmented version to: {output_filepath_with_bitrate} (Bitrate: {bitrate_kbps}k)")
+    print(
+        f"  -> Saving Opus augmented version to: {output_filepath_with_bitrate} (Bitrate: {bitrate_kbps}k)"
+    )
 
     # PyAV 需要 (n_channels, n_samples) 格式
     wav_ch_first = wav.T if wav.ndim > 1 else wav.reshape(1, -1)
@@ -81,19 +94,28 @@ def aac_aug_save(wav: np.ndarray, sr: int, output_filepath: str):
     对音频进行 AAC 压缩, 并直接保存到文件。
     使用 PyAV 实现, 比特率是随机的。
     """
+    try:
+        import av
+    except ImportError:
+        raise ImportError("av is required for aac_aug_save(). "
+                          "Please install it via `pip install av`.")
     # 随机选择一个比特率 (kbps)
     bitrate_kbps = random.choice([48, 64, 96, 128, 160, 192])
     # .m4a 是 AAC 更常用的文件后缀
-    output_filepath_with_bitrate = output_filepath.replace('.aac', f'_{bitrate_kbps}k.m4a')
+    output_filepath_with_bitrate = output_filepath.replace(
+        '.aac', f'_{bitrate_kbps}k.m4a')
 
-    print(f"  -> Saving AAC augmented version to: {output_filepath_with_bitrate} (Bitrate: {bitrate_kbps}k)")
+    print(
+        f"  -> Saving AAC augmented version to: {output_filepath_with_bitrate} (Bitrate: {bitrate_kbps}k)"
+    )
 
     # PyAV 需要 (n_channels, n_samples) 格式
     wav_ch_first = wav.T if wav.ndim > 1 else wav.reshape(1, -1)
     layout = 'stereo' if wav.ndim > 1 else 'mono'
 
     # 注意：format='adts' 是原始 AAC 流, 'mp4' 会创建 .m4a/.mp4 容器
-    with av.open(output_filepath_with_bitrate, mode='w', format='mp4') as container:
+    with av.open(output_filepath_with_bitrate, mode='w',
+                 format='mp4') as container:
         # 使用高质量的 fdk_aac 编码器
         stream = container.add_stream('libfdk_aac', rate=sr, layout=layout)
         stream.bit_rate = bitrate_kbps * 1000
@@ -111,13 +133,11 @@ def aac_aug_save(wav: np.ndarray, sr: int, output_filepath: str):
     print(f"     ... success.")
 
 
-def flac_encode_save(
-    wav: np.ndarray,
-    sr: int,
-    output_filepath: str,
-    compression_level: int = 5,
-    bits_per_sample=None
-):
+def flac_encode_save(wav: np.ndarray,
+                     sr: int,
+                     output_filepath: str,
+                     compression_level: int = 5,
+                     bits_per_sample=None):
     """
     使用 pyFLAC 将 NumPy 音频数组编码为 FLAC 文件并保存。
 
@@ -164,14 +184,14 @@ def flac_encode_save(
         else:
             wav_int = wav
     else:
-        raise ValueError(f"Unsupported bits_per_sample: {bits_per_sample}. Must be 16, 24, or 32.")
+        raise ValueError(
+            f"Unsupported bits_per_sample: {bits_per_sample}. Must be 16, 24, or 32."
+        )
 
     # --- 2. 初始化编码器 ---
-    encoder = Encoder(
-        sample_rate=sr,
-        bits_per_sample=bits_per_sample,
-        compression_level=compression_level
-    )
+    encoder = Encoder(sample_rate=sr,
+                      bits_per_sample=bits_per_sample,
+                      compression_level=compression_level)
 
     # --- 3. 处理数据并获取编码后的字节 ---
     # Encoder.process() 可以分块处理, 但对于中等长度的音频, 一次性处理更简单
@@ -196,8 +216,8 @@ def amr_wb_aug(wav, sr):
     # return ApplyCodec(encoder="libamr_wb", p=1.0)(samples, sample_rate)
     return sf.write('audio.amr', wav, sr, format='AMR', bitrate='192k')
 
-# Opus 编解码数据增强
 
+# Opus 编解码数据增强
 
 if __name__ == "__main__":
     wav_path = "/data/never/Desktop/kws_train/QA/wav_data/TIMIT.wav"