neverlib 0.2.3__py3-none-any.whl → 0.2.4__py3-none-any.whl

neverlib/.history/audio_aug/audio_aug_20250806010759.py

@@ -0,0 +1,140 @@
+# -*- coding:utf-8 -*-
+# Author:凌逆战 | Never
+# Date: 2024/9/27
+"""
+
+"""
+import random
+import numpy as np
+import soundfile as sf
+from scipy import signal
+from neverlib.utils import EPS
+
+
+def volume_norm(wav):
+    """
+    音量归一化
+    :param wav: (T,)
+    :return: (T,)
+    """
+    wav = wav / (np.max(np.abs(wav)) + 1e-8)
+    return wav
+
+
+def add_reverb(wav, rir, ratio=1, mode="same"):
+    """添加混响,
+    Args:
+        wav: [T, channel]
+        rir: [T, channel]
+        ratio:  0-1
+        mode: "same" for SE or "full" for kws
+    """
+    if random.random() < ratio:
+        wav = signal.fftconvolve(wav, rir, mode=mode)  # (28671, 3)
+        # note: 建议过完添加混响后再进行归一化, 否则可能会出现溢出
+        # wav = volume_norm(wav)
+    return wav
+
+
+def snr_aug_changeNoise(clean, noise, snr):
+    """ 
+    保持语音不变, 改变噪声的幅度
+    snr = 10 * log10(signal_power / k*noise_power) 
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)  # 纯净语音功率
+    noise_power = np.mean(noise ** 2)  # 噪声功率
+    noise_scale = np.sqrt(clean_power / (noise_power * 10 ** (snr / 10) + EPS))
+    noisy = clean + noise_scale * noise
+    return noisy, noise_scale
+
+
+def snr_aug_changeNoise(clean, noise, snr):
+    """ 
+    保持语音不变, 改变噪声的幅度
+    snr = 10 * log10(signal_power / k*noise_power) 
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)  # 纯净语音功率
+    noise_power = np.mean(noise ** 2)  # 噪声功率
+    snr_in = 10 * np.log10(clean_power / (noise_power + EPS) + EPS)
+    snr_gain = snr_in - snr
+    gain = 10 ** (snr_gain / 20)
+    noisy = clean + gain * noise
+    return noisy
+
+
+def snr_aug_changeClean(clean, noise, snr):
+    """ 
+    保持噪声不变, 改变语音的幅度
+    snr = 10 * log10(k*signal_power/ noise_power)
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)
+    noise_power = np.mean(noise ** 2)
+    clean_scale = np.sqrt(noise_power * 10 ** (snr / 10) / (clean_power + 1e-8))
+    noisy = clean * clean_scale + noise
+
+    return noisy, clean_scale
+
+
+def snr_aug_Interpolation(clean, noise, snr):
+    """
+    在已知clean_len<=noise_len的情况下
+    将clean插入到noise中的snr aug方法
+    Args:
+        clean: 语音
+        noise: 噪声
+        snr: snr=random.uniform(*snr_range)
+    """
+    clean_len, noise_len = clean.shape[0], noise.shape[0]
+    assert clean_len <= noise_len, f"clean_len must be less than noise_len."
+    noisy = noise.copy()
+    index = random.randint(0, noise_len - clean_len)
+    noise = noise[index:index + clean_len, :]
+    noisy_tmp, _ = snr_aug_changeClean(clean, noise, snr)
+    noisy[index:index + clean_len, :] = noisy_tmp
+    return noisy
+
+
+def get_audio_segments(wav_len, audio_path_list, sr=16000):
+    """
+    从音频列表中随机拼接指定长度音频
+    Args:
+        wav_len: 需要返回的音频长度
+        audio_path_list: 音频路径列表
+        sr: 采样率
+    Returns:返回指定长度的音频
+    """
+    audio_len = 0
+    wav_list = []
+    while audio_len < wav_len:
+        audio_path = random.choice(audio_path_list)
+        wav, wav_sr = sf.read(audio_path, always_2d=True, dtype='float32')
+        assert wav_sr == sr, f"音频采样率是{wav_sr}, 期望{sr}"
+        audio_len += len(wav)
+        wav_list.append(wav)
+    wav = np.concatenate(wav_list, axis=0)
+    if len(wav) > wav_len:
+        # 随机截取clean_len
+        start = random.randint(0, len(wav) - wav_len)
+        wav = wav[start:start + wav_len, :]
+    return wav
+
+
+def volume_aug(wav, range, rate, method="linmax"):
+    """音量增强 """
+    if random.random() < rate:
+        target_level = random.uniform(range[0], range[1])
+        if method == "dbrms":
+            wav_rms = (wav ** 2).mean() ** 0.5
+            scalar = 10 ** (target_level / 20) / (np.max(wav_rms) + EPS)
+        elif method == "linmax":
+            ipt_max = np.max(np.abs(wav))
+            # wav/wav_max*target_level=target_level_wav
+            # 处理后音频的 最大值就是target_level
+            scalar = target_level / (ipt_max + EPS)
+        else:
+            raise ValueError("method must be 'dbrms' or 'linmax'")
+        wav *= scalar
+    return wav

neverlib/.history/audio_aug/audio_aug_20250806010803.py

@@ -0,0 +1,140 @@
+# -*- coding:utf-8 -*-
+# Author:凌逆战 | Never
+# Date: 2024/9/27
+"""
+
+"""
+import random
+import numpy as np
+import soundfile as sf
+from scipy import signal
+from neverlib.utils import EPS
+
+
+def volume_norm(wav):
+    """
+    音量归一化
+    :param wav: (T,)
+    :return: (T,)
+    """
+    wav = wav / (np.max(np.abs(wav)) + 1e-8)
+    return wav
+
+
+def add_reverb(wav, rir, ratio=1, mode="same"):
+    """添加混响,
+    Args:
+        wav: [T, channel]
+        rir: [T, channel]
+        ratio:  0-1
+        mode: "same" for SE or "full" for kws
+    """
+    if random.random() < ratio:
+        wav = signal.fftconvolve(wav, rir, mode=mode)  # (28671, 3)
+        # note: 建议过完添加混响后再进行归一化, 否则可能会出现溢出
+        # wav = volume_norm(wav)
+    return wav
+
+
+def snr_aug_changeNoise(clean, noise, snr):
+    """ 
+    保持语音不变, 改变噪声的幅度
+    snr = 10 * log10(signal_power / k*noise_power) 
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)  # 纯净语音功率
+    noise_power = np.mean(noise ** 2)  # 噪声功率
+    noise_scale = np.sqrt(clean_power / (noise_power * 10 ** (snr / 10) + EPS))
+    noisy = clean + noise_scale * noise
+    return noisy, noise_scale
+
+
+def snr_aug_changeNoise(clean, noise, snr):
+    """ 
+    保持语音不变, 改变噪声的幅度
+    snr = 10 * log10(signal_power / k*noise_power) 
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)  # 纯净语音功率
+    noise_power = np.mean(noise ** 2)  # 噪声功率
+    snr_in = 10 * np.log10(clean_power / (noise_power + EPS) + EPS)
+    snr_gain = snr_in - snr
+    gain = 10 ** (snr_gain / 20)
+    noisy = clean + gain * noise
+    return noisy
+
+
+def snr_aug_changeClean(clean, noise, snr):
+    """ 
+    保持噪声不变, 改变语音的幅度
+    snr = 10 * log10(k*signal_power/ noise_power)
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)
+    noise_power = np.mean(noise ** 2)
+    clean_scale = np.sqrt(noise_power * 10 ** (snr / 10) / (clean_power + 1e-8))
+    noisy = clean * clean_scale + noise
+
+    return noisy, clean_scale
+
+
+def snr_aug_Interpolation(clean, noise, snr):
+    """
+    在已知clean_len<=noise_len的情况下
+    将clean插入到noise中的snr aug方法
+    Args:
+        clean: 语音
+        noise: 噪声
+        snr: snr=random.uniform(*snr_range)
+    """
+    clean_len, noise_len = clean.shape[0], noise.shape[0]
+    assert clean_len <= noise_len, f"clean_len must be less than noise_len."
+    noisy = noise.copy()
+    index = random.randint(0, noise_len - clean_len)
+    noise = noise[index:index + clean_len, :]
+    noisy_tmp, _ = snr_aug_changeClean(clean, noise, snr)
+    noisy[index:index + clean_len, :] = noisy_tmp
+    return noisy
+
+
+def get_audio_segments(wav_len, audio_path_list, sr=16000):
+    """
+    从音频列表中随机拼接指定长度音频
+    Args:
+        wav_len: 需要返回的音频长度
+        audio_path_list: 音频路径列表
+        sr: 采样率
+    Returns:返回指定长度的音频
+    """
+    audio_len = 0
+    wav_list = []
+    while audio_len < wav_len:
+        audio_path = random.choice(audio_path_list)
+        wav, wav_sr = sf.read(audio_path, always_2d=True, dtype='float32')
+        assert wav_sr == sr, f"音频采样率是{wav_sr}, 期望{sr}"
+        audio_len += len(wav)
+        wav_list.append(wav)
+    wav = np.concatenate(wav_list, axis=0)
+    if len(wav) > wav_len:
+        # 随机截取clean_len
+        start = random.randint(0, len(wav) - wav_len)
+        wav = wav[start:start + wav_len, :]
+    return wav
+
+
+def volume_aug(wav, range, rate, method="linmax"):
+    """音量增强 """
+    if random.random() < rate:
+        target_level = random.uniform(range[0], range[1])
+        if method == "dbrms":
+            wav_rms = (wav ** 2).mean() ** 0.5
+            scalar = 10 ** (target_level / 20) / (np.max(wav_rms) + EPS)
+        elif method == "linmax":
+            ipt_max = np.max(np.abs(wav))
+            # wav/wav_max*target_level=target_level_wav
+            # 处理后音频的 最大值就是target_level
+            scalar = target_level / (ipt_max + EPS)
+        else:
+            raise ValueError("method must be 'dbrms' or 'linmax'")
+        wav *= scalar
+    return wav

neverlib/.history/audio_aug/audio_aug_20250806010809.py

@@ -0,0 +1,140 @@
+# -*- coding:utf-8 -*-
+# Author:凌逆战 | Never
+# Date: 2024/9/27
+"""
+
+"""
+import random
+import numpy as np
+import soundfile as sf
+from scipy import signal
+from neverlib.utils import EPS
+
+
+def volume_norm(wav):
+    """
+    音量归一化
+    :param wav: (T,)
+    :return: (T,)
+    """
+    wav = wav / (np.max(np.abs(wav)) + 1e-8)
+    return wav
+
+
+def add_reverb(wav, rir, ratio=1, mode="same"):
+    """添加混响,
+    Args:
+        wav: [T, channel]
+        rir: [T, channel]
+        ratio:  0-1
+        mode: "same" for SE or "full" for kws
+    """
+    if random.random() < ratio:
+        wav = signal.fftconvolve(wav, rir, mode=mode)  # (28671, 3)
+        # note: 建议过完添加混响后再进行归一化, 否则可能会出现溢出
+        # wav = volume_norm(wav)
+    return wav
+
+
+def snr_aug_changeNoise(clean, noise, snr):
+    """ 
+    保持语音不变, 改变噪声的幅度
+    snr = 10 * log10(signal_power / k*noise_power) 
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)  # 纯净语音功率
+    noise_power = np.mean(noise ** 2)  # 噪声功率
+    noise_scale = np.sqrt(clean_power / (noise_power * 10 ** (snr / 10) + EPS))
+    noisy = clean + noise_scale * noise
+    return noisy, noise_scale
+
+
+def snr_aug_changeNoise_v2(clean, noise, snr):
+    """ 
+    保持语音不变, 改变噪声的幅度
+    snr = 10 * log10(signal_power / k*noise_power) 
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)  # 纯净语音功率
+    noise_power = np.mean(noise ** 2)  # 噪声功率
+    snr_in = 10 * np.log10(clean_power / (noise_power + EPS) + EPS)
+    snr_gain = snr_in - snr
+    gain = 10 ** (snr_gain / 20)
+    noisy = clean + gain * noise
+    return noisy
+
+
+def snr_aug_changeClean(clean, noise, snr):
+    """ 
+    保持噪声不变, 改变语音的幅度
+    snr = 10 * log10(k*signal_power/ noise_power)
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)
+    noise_power = np.mean(noise ** 2)
+    clean_scale = np.sqrt(noise_power * 10 ** (snr / 10) / (clean_power + 1e-8))
+    noisy = clean * clean_scale + noise
+
+    return noisy, clean_scale
+
+
+def snr_aug_Interpolation(clean, noise, snr):
+    """
+    在已知clean_len<=noise_len的情况下
+    将clean插入到noise中的snr aug方法
+    Args:
+        clean: 语音
+        noise: 噪声
+        snr: snr=random.uniform(*snr_range)
+    """
+    clean_len, noise_len = clean.shape[0], noise.shape[0]
+    assert clean_len <= noise_len, f"clean_len must be less than noise_len."
+    noisy = noise.copy()
+    index = random.randint(0, noise_len - clean_len)
+    noise = noise[index:index + clean_len, :]
+    noisy_tmp, _ = snr_aug_changeClean(clean, noise, snr)
+    noisy[index:index + clean_len, :] = noisy_tmp
+    return noisy
+
+
+def get_audio_segments(wav_len, audio_path_list, sr=16000):
+    """
+    从音频列表中随机拼接指定长度音频
+    Args:
+        wav_len: 需要返回的音频长度
+        audio_path_list: 音频路径列表
+        sr: 采样率
+    Returns:返回指定长度的音频
+    """
+    audio_len = 0
+    wav_list = []
+    while audio_len < wav_len:
+        audio_path = random.choice(audio_path_list)
+        wav, wav_sr = sf.read(audio_path, always_2d=True, dtype='float32')
+        assert wav_sr == sr, f"音频采样率是{wav_sr}, 期望{sr}"
+        audio_len += len(wav)
+        wav_list.append(wav)
+    wav = np.concatenate(wav_list, axis=0)
+    if len(wav) > wav_len:
+        # 随机截取clean_len
+        start = random.randint(0, len(wav) - wav_len)
+        wav = wav[start:start + wav_len, :]
+    return wav
+
+
+def volume_aug(wav, range, rate, method="linmax"):
+    """音量增强 """
+    if random.random() < rate:
+        target_level = random.uniform(range[0], range[1])
+        if method == "dbrms":
+            wav_rms = (wav ** 2).mean() ** 0.5
+            scalar = 10 ** (target_level / 20) / (np.max(wav_rms) + EPS)
+        elif method == "linmax":
+            ipt_max = np.max(np.abs(wav))
+            # wav/wav_max*target_level=target_level_wav
+            # 处理后音频的 最大值就是target_level
+            scalar = target_level / (ipt_max + EPS)
+        else:
+            raise ValueError("method must be 'dbrms' or 'linmax'")
+        wav *= scalar
+    return wav

neverlib/.history/audio_aug/audio_aug_20250806011108.py

@@ -0,0 +1,140 @@
+# -*- coding:utf-8 -*-
+# Author:凌逆战 | Never
+# Date: 2024/9/27
+"""
+
+"""
+import random
+import numpy as np
+import soundfile as sf
+from scipy import signal
+from neverlib.utils import EPS
+
+
+def volume_norm(wav):
+    """
+    音量归一化
+    :param wav: (T,)
+    :return: (T,)
+    """
+    wav = wav / (np.max(np.abs(wav)) + 1e-8)
+    return wav
+
+
+def add_reverb(wav, rir, ratio=1, mode="same"):
+    """添加混响,
+    Args:
+        wav: [T, channel]
+        rir: [T, channel]
+        ratio:  0-1
+        mode: "same" for SE or "full" for kws
+    """
+    if random.random() < ratio:
+        wav = signal.fftconvolve(wav, rir, mode=mode)  # (28671, 3)
+        # note: 建议过完添加混响后再进行归一化, 否则可能会出现溢出
+        # wav = volume_norm(wav)
+    return wav
+
+
+def snr_aug_changeNoise(clean, noise, snr):
+    """ 
+    保持语音不变, 改变噪声的幅度
+    snr = 10 * log10(signal_power / k*noise_power) 
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)  # 纯净语音功率
+    noise_power = np.mean(noise ** 2)  # 噪声功率
+    noise_scale = np.sqrt(clean_power / (noise_power * 10 ** (snr / 10) + EPS))
+    noisy = clean + noise_scale * noise
+    return noisy, noise_scale
+
+
+def snr_aug_changeNoise_v2(clean, noise, snr):
+    """ 
+    保持语音不变, 改变噪声的幅度
+    snr = 10 * log10(signal_power / k*noise_power) 
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)  # 纯净语音功率
+    noise_power = np.mean(noise ** 2)  # 噪声功率
+    snr_in = 10 * np.log10(clean_power / (noise_power + EPS) + EPS)
+    snr_gain = snr_in - snr
+    gain = 10 ** (snr_gain / 20)
+    noisy = clean + gain * noise
+    return noisy, gain
+
+
+def snr_aug_changeClean(clean, noise, snr):
+    """ 
+    保持噪声不变, 改变语音的幅度
+    snr = 10 * log10(k*signal_power/ noise_power)
+    """
+    assert clean.shape == noise.shape, "clean and noise must have the same shape"
+    clean_power = np.mean(clean ** 2)
+    noise_power = np.mean(noise ** 2)
+    clean_scale = np.sqrt(noise_power * 10 ** (snr / 10) / (clean_power + 1e-8))
+    noisy = clean * clean_scale + noise
+
+    return noisy, clean_scale
+
+
+def snr_aug_Interpolation(clean, noise, snr):
+    """
+    在已知clean_len<=noise_len的情况下
+    将clean插入到noise中的snr aug方法
+    Args:
+        clean: 语音
+        noise: 噪声
+        snr: snr=random.uniform(*snr_range)
+    """
+    clean_len, noise_len = clean.shape[0], noise.shape[0]
+    assert clean_len <= noise_len, f"clean_len must be less than noise_len."
+    noisy = noise.copy()
+    index = random.randint(0, noise_len - clean_len)
+    noise = noise[index:index + clean_len, :]
+    noisy_tmp, _ = snr_aug_changeClean(clean, noise, snr)
+    noisy[index:index + clean_len, :] = noisy_tmp
+    return noisy
+
+
+def get_audio_segments(wav_len, audio_path_list, sr=16000):
+    """
+    从音频列表中随机拼接指定长度音频
+    Args:
+        wav_len: 需要返回的音频长度
+        audio_path_list: 音频路径列表
+        sr: 采样率
+    Returns:返回指定长度的音频
+    """
+    audio_len = 0
+    wav_list = []
+    while audio_len < wav_len:
+        audio_path = random.choice(audio_path_list)
+        wav, wav_sr = sf.read(audio_path, always_2d=True, dtype='float32')
+        assert wav_sr == sr, f"音频采样率是{wav_sr}, 期望{sr}"
+        audio_len += len(wav)
+        wav_list.append(wav)
+    wav = np.concatenate(wav_list, axis=0)
+    if len(wav) > wav_len:
+        # 随机截取clean_len
+        start = random.randint(0, len(wav) - wav_len)
+        wav = wav[start:start + wav_len, :]
+    return wav
+
+
+def volume_aug(wav, range, rate, method="linmax"):
+    """音量增强 """
+    if random.random() < rate:
+        target_level = random.uniform(range[0], range[1])
+        if method == "dbrms":
+            wav_rms = (wav ** 2).mean() ** 0.5
+            scalar = 10 ** (target_level / 20) / (np.max(wav_rms) + EPS)
+        elif method == "linmax":
+            ipt_max = np.max(np.abs(wav))
+            # wav/wav_max*target_level=target_level_wav
+            # 处理后音频的 最大值就是target_level
+            scalar = target_level / (ipt_max + EPS)
+        else:
+            raise ValueError("method must be 'dbrms' or 'linmax'")
+        wav *= scalar
+    return wav

neverlib/.history/dataAnalyze/__init___20250806204125.py

@@ -0,0 +1,14 @@
+'''
+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。
+"""
+
+# 基础工具
+from neverlib.dataAnalyze.temporal_features import dB, peak_amplitude, rms_amplitude

neverlib/.history/dataAnalyze/__init___20250806204139.py

@@ -0,0 +1,14 @@
+'''
+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。
+"""
+
+# 基础工具
+from .dataAnalyze.temporal_features import dB, peak_amplitude, rms_amplitude

neverlib/.history/dataAnalyze/__init___20250806204159.py

@@ -0,0 +1,14 @@
+'''
+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。
+"""
+
+# 基础工具
+from neverlib.dataAnalyze.temporal_features import dB, peak_amplitude, rms_amplitude

neverlib/.history/filter/__init___20250820103351.py

@@ -0,0 +1,70 @@
+'''
+Author: 凌逆战 | Never
+Date: 2025-03-17 19:23:33
+Description: 
+'''
+"""
+节省路径
+from neverlib.filter import common
+如果没有用户必须完整路径
+from neverlib.filter.common import *
+"""
+from .common import *
+from .core import *
+from .biquad import *
+
+def __getattr__(name):
+    """延迟导入机制 - 只在用户实际使用时才导入需要额外依赖的模块"""
+    if name in ['compute_frequency_eq']:
+        # 需要 librosa 库的频谱直接补偿方法
+        try:
+            from .AudoEQ.auto_eq_spectral_direct import compute_frequency_eq
+            return compute_frequency_eq
+        except ImportError as e:
+            raise ImportError(f"使用 {name} 需要安装 librosa: pip install librosa") from e
+            
+    elif name in ['get_filter_function', 'match_frequency_response', 'plot_spectra_comparison']:
+        # 差分进化优化方法
+        try:
+            from .AudoEQ.auto_eq_de import get_filter_function, match_frequency_response, plot_spectra_comparison
+            if name == 'get_filter_function':
+                return get_filter_function
+            elif name == 'match_frequency_response':
+                return match_frequency_response
+            elif name == 'plot_spectra_comparison':
+                return plot_spectra_comparison
+        except ImportError as e:
+            raise ImportError(f"使用 {name} 需要安装相应依赖，详见 AudoEQ/README.md") from e
+            
+    elif name in ['individual_creator', 'get_magnitude_spectrum_db', 'get_single_filter_freq_response_db_from_coeffs', 
+                  'get_combined_eq_response_db', 'evaluate_individual', 'custom_mutate']:
+        # 需要 deap 库的遗传算法基础版
+        try:
+            from .AudoEQ.auto_eq_ga_basic import (
+                individual_creator, get_magnitude_spectrum_db, get_single_filter_freq_response_db_from_coeffs,
+                get_combined_eq_response_db, evaluate_individual, custom_mutate
+            )
+            if name == 'individual_creator':
+                return individual_creator
+            elif name == 'get_magnitude_spectrum_db':
+                return get_magnitude_spectrum_db
+            elif name == 'get_single_filter_freq_response_db_from_coeffs':
+                return get_single_filter_freq_response_db_from_coeffs
+            elif name == 'get_combined_eq_response_db':
+                return get_combined_eq_response_db
+            elif name == 'evaluate_individual':
+                return evaluate_individual
+            elif name == 'custom_mutate':
+                return custom_mutate
+        except ImportError as e:
+            raise ImportError(f"使用 {name} 需要安装 deap: pip install deap") from e
+            
+    elif name == 'EQOptimizer':
+        # 需要 deap 库的遗传算法高级版
+        try:
+            from .AudoEQ.auto_eq_ga_advanced import EQOptimizer
+            return EQOptimizer
+        except ImportError as e:
+            raise ImportError(f"使用 {name} 需要安装 deap: pip install deap") from e
+    
+    raise AttributeError(f"module '{__name__}' has no attribute '{name}'")