neverlib 0.2.6__py3-none-any.whl → 0.2.8__py3-none-any.whl

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"

neverlib/audio_aug/coder_aug2.py

@@ -3,23 +3,23 @@ Author: 凌逆战 | Never
 Date: 2025-07-29 17:57:26
 Description: 
 '''
-import numpy as np
-import librosa
-import soundfile as sf
-import subprocess
 import os
 import random
-import io
+import soundfile as sf
+import subprocess
 
 
 def check_codec_available(codec):
     """检查指定的编解码器是否在FFmpeg中可用"""
     try:
-        result = subprocess.run(['ffmpeg', '-encoders'], capture_output=True, text=True)
+        result = subprocess.run(['ffmpeg', '-encoders'],
+                                capture_output=True,
+                                text=True)
         return codec in result.stdout
     except:
         return False
 
+
 def apply_codec_distortion(wav, sr, codec='libopus', bitrate='24k'):
     """
     使用 FFmpeg 对音频应用指定的编解码器和码率, 以模拟有损压缩失真。
@@ -56,12 +56,15 @@ def apply_codec_distortion(wav, sr, codec='libopus', bitrate='24k'):
         sf.write(input_filename, wav, sr)
 
         # 2. 构建 FFmpeg 命令
-        command = ['ffmpeg', '-y', '-i', input_filename, '-c:a', codec, '-b:a', bitrate]
+        command = [
+            'ffmpeg', '-y', '-i', input_filename, '-c:a', codec, '-b:a',
+            bitrate
+        ]
 
         # 3. 为 AMR 编解码器添加重采样参数
         if codec in ['amr_nb', 'amr_wb']:
             command.extend(['-ar', '8000'])  # AMR-NB 需要 8kHz 采样率
-        
+
         # 4. 为 AAC 指定输出格式 (移除 -f adts, 使用 MP4 容器)
         # if codec == 'aac':
         #     command.extend(['-f', 'adts'])
@@ -69,13 +72,18 @@ def apply_codec_distortion(wav, sr, codec='libopus', bitrate='24k'):
         command.append(output_filename)
 
         # 执行命令, 并隐藏输出
-        subprocess.run(command, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+        subprocess.run(command,
+                       check=True,
+                       stdout=subprocess.PIPE,
+                       stderr=subprocess.PIPE)
 
         # 4. 将编码后的文件转换回 WAV 格式以便读取
         wav_output = f"temp_final_{codec}_{bitrate}.wav"
-        subprocess.run(['ffmpeg', '-y', '-i', output_filename, wav_output], 
-                      check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-        
+        subprocess.run(['ffmpeg', '-y', '-i', output_filename, wav_output],
+                       check=True,
+                       stdout=subprocess.PIPE,
+                       stderr=subprocess.PIPE)
+
         # 5. 从 WAV 文件读回失真的音频
         samples = sf.read(wav_output)[0]
 
@@ -87,32 +95,41 @@ def apply_codec_distortion(wav, sr, codec='libopus', bitrate='24k'):
         return wav
     finally:
         # 6. 清理临时文件
-        for temp_file in [input_filename, output_filename, f"temp_final_{codec}_{bitrate}.wav"]:
+        for temp_file in [
+                input_filename, output_filename,
+                f"temp_final_{codec}_{bitrate}.wav"
+        ]:
             if os.path.exists(temp_file):
                 os.remove(temp_file)
 
 
-# --- 使用示例 ---
-wav_path = "/data/never/Desktop/kws_train/QA/wav_data/TIMIT.wav"
-wav, wav_sr = sf.read(wav_path, always_2d=True)
-
-# 1. 模拟 Opus 编解码器（常用于VoIP, WebRTC）
-print("应用 Opus 编解码器失真...")
-opus_wav = apply_codec_distortion(wav, wav_sr, codec='libopus', bitrate='24k')
-sf.write('augmented_opus.wav', opus_wav, wav_sr)
-
-# 2. 模拟 AAC 编解码器（常用于流媒体, Apple设备）
-print("应用 AAC 编解码器失真...")
-aac_wav = apply_codec_distortion(wav, wav_sr, codec='aac', bitrate='64k')
-sf.write('augmented_aac.wav', aac_wav, wav_sr)
-
-# 3. 模拟 AMR-NB 编解码器（常用于传统移动通信）
-# AMR-NB 的码率是固定的几个值之一
-amr_bitrates = ['4.75k', '5.15k', '5.9k', '6.7k', '7.4k', '7.95k', '10.2k', '12.2k']
-chosen_amr_bitrate = random.choice(amr_bitrates)
-print(f"应用 AMR-NB @ {chosen_amr_bitrate} 编解码器失真...")
-amr_wav = apply_codec_distortion(wav, wav_sr, codec='amr_nb', bitrate=chosen_amr_bitrate)
-# 注意：AMR通常是8kHz采样, librosa加载时会自动重采样, 这里我们保持原始sr
-sf.write('augmented_amr.wav', amr_wav, wav_sr)
-
-print("所有编解码器增强完成！")
+if __name__ == "__main__":
+    # --- 使用示例 ---
+    wav_path = "/data/never/Desktop/kws_train/QA/wav_data/TIMIT.wav"
+    wav, wav_sr = sf.read(wav_path, always_2d=True)
+
+    # 1. 模拟 Opus 编解码器（常用于VoIP, WebRTC）
+    print("应用 Opus 编解码器失真...")
+    opus_wav = apply_codec_distortion(wav, wav_sr, codec='libopus', bitrate='24k')
+    sf.write('augmented_opus.wav', opus_wav, wav_sr)
+
+    # 2. 模拟 AAC 编解码器（常用于流媒体, Apple设备）
+    print("应用 AAC 编解码器失真...")
+    aac_wav = apply_codec_distortion(wav, wav_sr, codec='aac', bitrate='64k')
+    sf.write('augmented_aac.wav', aac_wav, wav_sr)
+
+    # 3. 模拟 AMR-NB 编解码器（常用于传统移动通信）
+    # AMR-NB 的码率是固定的几个值之一
+    amr_bitrates = [
+        '4.75k', '5.15k', '5.9k', '6.7k', '7.4k', '7.95k', '10.2k', '12.2k'
+    ]
+    chosen_amr_bitrate = random.choice(amr_bitrates)
+    print(f"应用 AMR-NB @ {chosen_amr_bitrate} 编解码器失真...")
+    amr_wav = apply_codec_distortion(wav,
+                                     wav_sr,
+                                     codec='amr_nb',
+                                     bitrate=chosen_amr_bitrate)
+    # 注意：AMR通常是8kHz采样, librosa加载时会自动重采样, 这里我们保持原始sr
+    sf.write('augmented_amr.wav', amr_wav, wav_sr)
+
+    print("所有编解码器增强完成！")

neverlib/audio_aug/loss_packet_aug.py

@@ -18,7 +18,6 @@ Description: 丢包数据增强
 “置零”是在深度学习框架下对“真正丢弃”问题的一种高效、可解的数学建模。 我们牺牲了一点点物理上的真实性, 换来了模型训练的可行性和高效性。
 '''
 import numpy as np
-import random
 import soundfile as sf
 
 
@@ -93,11 +92,12 @@ def simulate_packet_loss_vectorized(
     return damaged_wav
 
 
-# 生成一个白噪声
-white_noise = np.random.randn(100000).astype(np.float32)
+if __name__ == "__main__":
+    # 生成一个白噪声
+    white_noise = np.random.randn(100000).astype(np.float32)
 
-# 生成一个损坏的音频
-damaged_audio = simulate_packet_loss_vectorized(white_noise, 16000, loss_rate=0.1, burst_prob=0.5)
+    # 生成一个损坏的音频
+    damaged_audio = simulate_packet_loss_vectorized(white_noise, 16000, loss_rate=0.1, burst_prob=0.5)
 
-# 保存音频
-sf.write("damaged_audio.wav", damaged_audio, 16000)
+    # 保存音频
+    sf.write("damaged_audio.wav", damaged_audio, 16000)

neverlib/audio_aug/quant_aug.py

@@ -3,13 +3,8 @@ Author: 凌逆战 | Never
 Date: 2025-03-26 22:13:21
 Description: 
 '''
-import os
-import sys
 import numpy as np
 import soundfile as sf
-import librosa
-
-print(librosa.__version__)
 
 
 def apply_uniform_quantization(wav, bit_depth=8):
@@ -47,6 +42,13 @@ def apply_mulaw_quantization(wav, bit_depth=8):
     wav (np.ndarray): 输入音频。
     bit_depth (int): 目标模拟的位深度。
     """
+    try:
+        import librosa
+    except ImportError:
+        raise ImportError(
+            "librosa is required for apply_mulaw_quantization(). "
+            "Please install it via `pip install librosa`."
+        )
     # mu 的值决定了量化级别的数量 (mu + 1)
     mu = 2**bit_depth - 1
 
@@ -61,18 +63,18 @@ def apply_mulaw_quantization(wav, bit_depth=8):
     return expanded_wav
 
 
-# --- 使用示例 ---
-wav_path = "/data/never/Desktop/kws_train/QA/wav_data/TIMIT.wav"
-wav, wav_sr = sf.read(wav_path, always_2d=True)
-
-# 模拟一个 8-bit 的老式数字音频设备
-y_quantized_8bit = apply_uniform_quantization(wav, bit_depth=8)
-sf.write('augmented_quantized_8bit.wav', y_quantized_8bit, wav_sr)
+if __name__ == "__main__":
+    # --- 使用示例 ---
+    wav_path = "/data/never/Desktop/kws_train/QA/wav_data/TIMIT.wav"
+    wav, wav_sr = sf.read(wav_path, always_2d=True)
 
-# 模拟一个更差的 4-bit 设备
-y_quantized_4bit = apply_uniform_quantization(wav, bit_depth=4)
-sf.write('augmented_quantized_4bit.wav', y_quantized_4bit, wav_sr)
+    # 模拟一个 8-bit 的老式数字音频设备
+    y_quantized_8bit = apply_uniform_quantization(wav, bit_depth=8)
+    sf.write('augmented_quantized_8bit.wav', y_quantized_8bit, wav_sr)
 
+    # 模拟一个更差的 4-bit 设备
+    y_quantized_4bit = apply_uniform_quantization(wav, bit_depth=4)
+    sf.write('augmented_quantized_4bit.wav', y_quantized_4bit, wav_sr)
 
-y_q = apply_mulaw_quantization(wav, bit_depth=8)
-sf.write('augmented_mulaw_8bit.wav', y_q, wav_sr)
+    y_q = apply_mulaw_quantization(wav, bit_depth=8)
+    sf.write('augmented_mulaw_8bit.wav', y_q, wav_sr)

neverlib/data_analyze/__init__.py

@@ -1,20 +1,69 @@
-'''
-Author: 凌逆战 | Never
-Date: 2025-08-06 00:56:39
-Description: 
-'''
-"""
-音频数据分析模块
-Audio Data Analysis Module
-
-提供完整的音频数据分析功能, 包括特征提取、质量评估、统计分析和可视化等。
-"""
-from lazy_loader import attach
-
-__getattr__, __dir__, __all__ = attach(
-    __name__,
-    submodules=["temporal_features"],
-    submod_attrs={
-        "temporal_features": ["dB", "peak_amplitude", "rms_amplitude"],
-    },
-)
+# This file is auto-generated. Do NOT edit manually.
+# Generated by neverlib.QA.gen_init
+
+from typing import TYPE_CHECKING, Any
+
+if TYPE_CHECKING:
+    # 仅在类型检查时导入，提供IDE补全支持
+    from .dataset_analyzer import AudioFileInfo, DatasetAnalyzer, analyze_audio_dataset
+    from .quality_metrics import QualityAnalyzer, audio_health_check, comprehensive_quality_assessment
+    from .rms_distrubution import get_rms_vad
+    from .spectral_analysis import SpectralAnalyzer, compute_spectral_features, frequency_domain_stats
+    from .statistics import AudioStatistics, compare_datasets, quick_audio_stats
+    from .temporal_features import dB, dc_offset, max_rms_amplitude, mean_rms_amplitude, min_rms_amplitude, peak_amplitude, rms_amplitude, short_time_energy, zero_crossing_rate
+    from .visualization import AudioVisualizer, create_analysis_dashboard, plot_dataset_overview
+
+# 运行时使用懒加载
+from lazy_loader import attach
+
+__getattr__, __dir__, __all__ = attach(
+    __name__,
+    submodules=[
+        "dataset_analyzer",
+        "quality_metrics",
+        "rms_distrubution",
+        "spectral_analysis",
+        "statistics",
+        "temporal_features",
+        "visualization",
+    ],
+    submod_attrs={
+        "dataset_analyzer": ['AudioFileInfo', 'DatasetAnalyzer', 'analyze_audio_dataset'],
+        "quality_metrics": ['QualityAnalyzer', 'audio_health_check', 'comprehensive_quality_assessment'],
+        "rms_distrubution": ['get_rms_vad'],
+        "spectral_analysis": ['SpectralAnalyzer', 'compute_spectral_features', 'frequency_domain_stats'],
+        "statistics": ['AudioStatistics', 'compare_datasets', 'quick_audio_stats'],
+        "temporal_features": ['dB', 'dc_offset', 'max_rms_amplitude', 'mean_rms_amplitude', 'min_rms_amplitude', 'peak_amplitude', 'rms_amplitude', 'short_time_energy', 'zero_crossing_rate'],
+        "visualization": ['AudioVisualizer', 'create_analysis_dashboard', 'plot_dataset_overview'],
+    }
+)
+
+# 显式声明 __all__ 以便 IDE 识别
+if TYPE_CHECKING:
+    __all__ = [
+        'AudioFileInfo',
+        'DatasetAnalyzer',
+        'analyze_audio_dataset',
+        'QualityAnalyzer',
+        'audio_health_check',
+        'comprehensive_quality_assessment',
+        'get_rms_vad',
+        'SpectralAnalyzer',
+        'compute_spectral_features',
+        'frequency_domain_stats',
+        'AudioStatistics',
+        'compare_datasets',
+        'quick_audio_stats',
+        'dB',
+        'dc_offset',
+        'max_rms_amplitude',
+        'mean_rms_amplitude',
+        'min_rms_amplitude',
+        'peak_amplitude',
+        'rms_amplitude',
+        'short_time_energy',
+        'zero_crossing_rate',
+        'AudioVisualizer',
+        'create_analysis_dashboard',
+        'plot_dataset_overview',
+    ]