neverlib 0.2.8__tar.gz → 0.2.9__tar.gz

{neverlib-0.2.8/neverlib.egg-info → neverlib-0.2.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: neverlib
-Version: 0.2.8
+Version: 0.2.9
 Summary: A successful sign for python setup
 Author-email: "Never.Ling" <1786088386@qq.com>
 License: MIT
@@ -47,6 +47,14 @@ Requires-Dist: neverlib[vad]; extra == "all"
 Requires-Dist: neverlib[gpu]; extra == "all"
 Dynamic: license-file
 
+<!--
+ * @Author: never.ling never.ling@anker-in.com
+ * @Date: 2025-11-04 17:10:57
+ * @LastEditors: never.ling never.ling@anker-in.com
+ * @LastEditTime: 2025-12-15 11:47:03
+ * @FilePath: /neverlib/README.md
+ * @Description: 这是默认设置,请设置`customMade`, 打开koroFileHeader查看配置 进行设置: https://github.com/OBKoro1/koro1FileHeader/wiki/%E9%85%8D%E7%BD%AE
+-->
 # NeverLib
 
 一个用于音频处理和VAD（语音活动检测）的Python工具库。
@@ -113,6 +121,12 @@ from neverlib import utils
 from neverlib.message import seed_QQEmail
 ```
 
+## TODO List
+- [ ]  客观指标添加 [DNSMOSPro ](https://github.com/fcumlin/DNSMOSPro)
+- [ ]  github界面的自我介绍
+- [ ]  https://github.com/nathanhaigh/parallel-rsync
+
+
 ## 许可证
 
 本项目采用MIT许可证。详情请参阅LICENSE文件。

{neverlib-0.2.8 → neverlib-0.2.9}/README.md

@@ -1,3 +1,11 @@
+<!--
+ * @Author: never.ling never.ling@anker-in.com
+ * @Date: 2025-11-04 17:10:57
+ * @LastEditors: never.ling never.ling@anker-in.com
+ * @LastEditTime: 2025-12-15 11:47:03
+ * @FilePath: /neverlib/README.md
+ * @Description: 这是默认设置,请设置`customMade`, 打开koroFileHeader查看配置 进行设置: https://github.com/OBKoro1/koro1FileHeader/wiki/%E9%85%8D%E7%BD%AE
+-->
 # NeverLib
 
 一个用于音频处理和VAD（语音活动检测）的Python工具库。
@@ -64,6 +72,12 @@ from neverlib import utils
 from neverlib.message import seed_QQEmail
 ```
 
+## TODO List
+- [ ]  客观指标添加 [DNSMOSPro ](https://github.com/fcumlin/DNSMOSPro)
+- [ ]  github界面的自我介绍
+- [ ]  https://github.com/nathanhaigh/parallel-rsync
+
+
 ## 许可证
 
 本项目采用MIT许可证。详情请参阅LICENSE文件。

neverlib-0.2.9/neverlib/QA/ImpactNoiseRejection.py

@@ -0,0 +1,119 @@
+"""VPU 冲击声检测 + 可视化（轻量版）。"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import List, Tuple
+
+import matplotlib.pyplot as plt
+import numpy as np
+import soundfile as sf
+
+# 基本参数，可按需修改
+AUDIO_PATH = Path(__file__).resolve().parent / "out.wav"
+CHANNEL_INDEX = 4
+DC_WINDOW_MS = 40.0
+SHORT_MS = 6.0
+LONG_MS = 60.0
+RATIO_THRESHOLD = 3.0
+MIN_DURATION_MS = 8.0
+MIN_GAP_MS = 35.0
+
+
+def moving_average(x: np.ndarray, window: int) -> np.ndarray:
+    if window <= 1:
+        return x
+    kernel = np.ones(window, dtype=np.float64) / window
+    return np.convolve(x, kernel, mode="same")
+
+
+def preprocess(signal: np.ndarray, sr: int) -> np.ndarray:
+    dc_window = max(int(sr * DC_WINDOW_MS / 1000.0), 1)
+    return signal - moving_average(signal, dc_window)
+
+
+def ratio_score(signal: np.ndarray, sr: int) -> np.ndarray:
+    short = max(int(sr * SHORT_MS / 1000.0), 1)
+    long = max(int(sr * LONG_MS / 1000.0), 1)
+    short_env = moving_average(np.abs(signal), short)
+    long_env = moving_average(np.abs(signal), long)
+    return short_env / np.maximum(long_env, 1e-6)
+
+
+def detect_events(mask: np.ndarray, sr: int) -> List[Tuple[float, float]]:
+    min_samples = max(int(sr * MIN_DURATION_MS / 1000.0), 1)
+    gap_samples = max(int(sr * MIN_GAP_MS / 1000.0), 1)
+
+    events: List[Tuple[int, int]] = []
+    start = -1
+    for idx, active in enumerate(mask):
+        if active and start < 0:
+            start = idx
+        elif not active and start >= 0:
+            if idx - start >= min_samples:
+                events = _append_or_merge(events, start, idx, gap_samples)
+            start = -1
+
+    if start >= 0 and len(mask) - start >= min_samples:
+        events = _append_or_merge(events, start, len(mask), gap_samples)
+
+    return [(s / sr, e / sr) for s, e in events]
+
+
+def _append_or_merge(events: List[Tuple[int, int]], start: int, end: int, gap: int) -> List[Tuple[int, int]]:
+    if events and start - events[-1][1] <= gap:
+        previous_start, _ = events[-1]
+        events[-1] = (previous_start, end)
+    else:
+        events.append((start, end))
+    return events
+
+
+def plot_detection(time: np.ndarray, signal: np.ndarray, score: np.ndarray, events: List[Tuple[float, float]]) -> None:
+    fig, axes = plt.subplots(2, 1, figsize=(12, 6), sharex=True)
+
+    axes[0].plot(time, signal, color="steelblue", linewidth=0.8)
+    axes[0].set_ylabel("Amplitude")
+    axes[0].set_title("VPU 信号与冲击事件")
+
+    axes[1].plot(time, score, color="crimson", linewidth=0.8)
+    axes[1].axhline(RATIO_THRESHOLD, color="gray", linestyle="--", linewidth=0.8, label="阈值")
+    axes[1].set_ylabel("短/长包络比")
+    axes[1].set_xlabel("时间 (s)")
+    axes[1].legend(loc="upper right")
+
+    for axis in axes:
+        for start, end in events:
+            axis.axvspan(start, end, color="lime", alpha=0.25)
+        axis.grid(alpha=0.2)
+
+    fig.tight_layout()
+    plt.savefig('./impact_noise_rejection.png')
+
+
+def main() -> None:
+    if not AUDIO_PATH.exists():
+        raise FileNotFoundError(f"找不到音频文件: {AUDIO_PATH}")
+
+    data, sr = sf.read(AUDIO_PATH, always_2d=True)
+    if CHANNEL_INDEX >= data.shape[1]:
+        raise ValueError(f"音频仅有 {data.shape[1]} 通道，无法访问 VPU 通道 {CHANNEL_INDEX}")
+
+    signal = data[:, CHANNEL_INDEX].astype(np.float64)
+    signal = preprocess(signal, sr)
+    score = ratio_score(signal, sr)
+    events = detect_events(score > RATIO_THRESHOLD, sr)
+
+    if events:
+        print("检测到的冲击事件：")
+        for start, end in events:
+            print(f"  起始 {start:.3f}s  结束 {end:.3f}s  持续 {(end-start)*1000:.1f}ms")
+    else:
+        print("未检测到明显的冲击事件。")
+
+    time_axis = np.arange(len(signal)) / sr
+    plot_detection(time_axis, signal, score, events)
+
+
+if __name__ == "__main__":
+    main()

neverlib-0.2.9/neverlib/audio_aug/README.md

@@ -0,0 +1,3 @@
+https://github.com/iver56/audiomentations
+https://github.com/gfdb/wav2aug
+https://github.com/documentation-ai/DocumentationAI-Docs

neverlib-0.2.9/neverlib/audio_aug/__init__.py

@@ -0,0 +1,28 @@
+# 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",
+        "TFDrop",
+        "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-0.2.9/neverlib/data_analyze/__init__.py

@@ -0,0 +1,25 @@
+# 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=[
+        "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'],
+    }
+)

neverlib-0.2.9/neverlib/filter/__init__.py

@@ -0,0 +1,17 @@
+# 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=[
+        "biquad",
+        "common",
+        "core",
+    ],
+    submod_attrs={
+        "biquad": ['BiquadFilter'],
+        "common": ['HPFilter', 'HPFilter_torch', 'LPFilter'],
+        "core": ['EQFilter', 'EQ_test', 'eq_process', 'eq_process_test'],
+    }
+)

neverlib-0.2.9/neverlib/filter/auto_eq/__init__.py

@@ -0,0 +1,19 @@
+# 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=[
+        "de_eq",
+        "freq_eq",
+        "ga_eq_advanced",
+        "ga_eq_basic",
+    ],
+    submod_attrs={
+        "de_eq": ['get_filter_function', 'match_frequency_response', 'plot_spectra_comparison'],
+        "freq_eq": ['get_freq_eq'],
+        "ga_eq_advanced": ['EQConfig', 'EQOptimizer', 'load_config_from_yaml', 'main'],
+        "ga_eq_basic": ['custom_mutate', 'evaluate_individual', 'generate_active_gene', 'generate_dbgain_gene', 'generate_fc_gene', 'generate_q_gene', 'generate_type_gene', 'get_combined_eq_response_db', 'get_magnitude_spectrum_db', 'get_single_filter_freq_response_db_from_coeffs', 'individual_creator', 'main_ga'],
+    }
+)

{neverlib-0.2.8 → neverlib-0.2.9}/neverlib/filter/core.py

@@ -322,9 +322,8 @@ def EQ_test():
     from scipy import signal
 
     fs = 16000
-
-    # b, a = LowshelfFilter(1500, 15, Q=0.5)
-    b, a = PeakingFilter(1500, 5, Q=0.5)
+    eq = EQFilter(fs)
+    b, a = eq.LowshelfFilter(4000, 6, Q=0.7)
     print(b)
     print(a)
 
@@ -335,5 +334,9 @@ def EQ_test():
     plt.ylabel('Amplitude [dB]')
     plt.grid(which='both', axis='both')  # 显示网格
     # 画红色的垂直线, 标记截止频率
-    plt.xscale('log')  # x轴对数化
-    plt.show()
+    # plt.xscale('log')  # x轴对数化
+    plt.savefig('./lowshelf_freq_response.png')
+
+
+if __name__ == "__main__":
+    EQ_test()

neverlib-0.2.9/neverlib/metrics/README.md

@@ -0,0 +1,35 @@
+https://github.com/ftshijt/speech_evaluation?tab=readme-ov-file
+https://github.com/Amir-Ivry/MAPSS-measures
+https://github.com/espnet/espnet/blob/f517b63fe5defb865e2067b11c2dfc516150868d/utils/eval-source-separation.py#L361
+
+
+
+
+
+F0 Correlation
+
+F0 Root Mean Square Error
+sdr
+isr
+sir
+sar
+perm
+
+
+Convolutional scale-invariant signal-to-distortion ratio (CI-SDR) [x]
+Scale-invariant signal-to-noise ratio (SI-SNR) [x]
+Speech BERT Score
+Discrete Speech BLEU Score [x]
+Discrete Speech Token Edit Distance
+
+
+
+
+STOI
+ESTOI
+PESQ
+
+
+音频特征：
+https://github.com/libAudioFlux/audioFlux
+https://github.com/csteinmetz1/pyloudnorm

neverlib-0.2.9/neverlib/metrics/__init__.py

@@ -0,0 +1,23 @@
+# 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=[
+        "dnsmos",
+        "lpc_lsp",
+        "snr",
+        "spec",
+        "test_pesq",
+        "time",
+    ],
+    submod_attrs={
+        "dnsmos": ['ComputeScore'],
+        "lpc_lsp": ['framing', 'lpc_lsp_distance', 'lpc_to_lsp'],
+        "snr": ['get_snr', 'get_snr_from_noisy', 'psnr', 'seg_snr', 'si_sdr'],
+        "spec": ['lsd', 'mcd', 'sd'],
+        "test_pesq": ['mos2pesq', 'pesq2mos'],
+        "time": ['dc_offset', 'mean_rms_amplitude', 'peak_amplitude', 'rms_amplitude'],
+    }
+)

neverlib-0.2.9/neverlib/signal_gen/babble_noise_generate.py

@@ -0,0 +1,113 @@
+# -*- coding:utf-8 -*-
+# Author:凌逆战 | Never
+# Date: 2024/6/14
+"""
+babble噪声生成
+"""
+import os
+import torch
+import random
+import numpy as np
+import soundfile as sf
+import pickle
+
+from neverlib.utils import get_path_list
+from vad import EnergyVAD
+from utils.audio_aug import volume_aug, add_reverb
+from joblib import Parallel, delayed
+from multiprocessing import Pool
+
+vad = EnergyVAD(
+    sample_rate=16000,
+    frame_length=25,  # in milliseconds
+    frame_shift=20,  # in milliseconds
+    energy_threshold=0.05,  # you may need to adjust this value
+    pre_emphasis=0.95,
+)  # default values are used here
+
+
+def get_one_babble(speech_path_list, rir_path_list):
+    sub_len = 0
+    sub_list = []
+    while sub_len < sentence_len:
+        speech_path = random.choice(speech_path_list)
+        rir_path = random.choice(rir_path_list)
+        # 加rir
+        with open(rir_path, 'rb') as f:
+            # arrayCoor (2,3) (num_mic, 3(x,y,z))
+            # sPos (4,3) (num_source, 3(x,y,z))
+            [arrayCoor, sPos, sAziPth, rir, L, rt, c, mtype, n, fs] = pickle.load(f)
+            rir = np.array(rir)  # (souece, mic, T) (4, 3, 2668)
+            rir_idx = np.random.randint(0, rir.shape[0])  # 随机选择一个source
+            sPos = sPos[rir_idx]  # (3,) (x,y,z)
+            DistMic2Source_1 = np.sqrt(np.sum((arrayCoor[0] - sPos) ** 2))
+            DistMic2Source_2 = np.sqrt(np.sum((arrayCoor[1] - sPos) ** 2))
+            if DistMic2Source_1 < 2 or DistMic2Source_2 < 2:
+                continue
+            rir = rir[rir_idx]
+
+        wav, _ = sf.read(speech_path, dtype='float32', always_2d=True)
+        assert wav.shape[-1] == 1, f"音频应该为单通道"
+        if len(wav) > sentence_len:
+            idx = random.randint(0, len(wav) - sentence_len)
+            wav = wav[idx:idx + sentence_len]
+
+        wav = wav / np.abs(wav).max()  # 归一化
+        wav = vad.apply_vad(wav.T).T
+
+        wav = add_reverb(wav, rir, add_rir_ratio=1)  # 双麦wav
+        # 音量增强
+        rms = np.sqrt(np.mean(wav ** 2))
+        wav *= random.uniform(0.5, 1.0) / rms
+
+        sub_len += len(wav)
+        sub_list.append(wav)
+
+    sub_babble = np.concatenate(sub_list, axis=0)
+    if sub_len >= sentence_len:
+        sub_babble = sub_babble[:sentence_len]
+    return sub_babble
+
+
+def babble_noise_generate(i, speech_path_list, rir_path_list):
+    source_num = random.randint(15, 25)
+    babble_noise = np.zeros((sentence_len, 2))
+    for _ in range(source_num):
+        sub_babble = get_one_babble(speech_path_list, rir_path_list)
+        babble_noise += sub_babble
+
+    babble_path = os.path.join(target_dir, f"babble_noise_{i}.wav")
+    print(babble_path)
+    babble_noise = babble_noise / np.max(np.abs(babble_noise)) * 0.5
+    sf.write(babble_path, babble_noise, sr)
+
+
+if __name__ == "__main__":
+    speech_dir = "/data/never/kws_data/BigDataset/speech"
+    rir_dir = "/data/never/RIR/2mic_Line_r50mm_fs16k_4source_t60mix08_maxdist8m_2w"
+    target_dir = "/data/never/SE_dataset/babble_noise10"
+    # if os.path.exists(target_dir):
+    #     os.system(f"rm -rf {target_dir}")
+    #     os.makedirs(target_dir, exist_ok=True)
+    # else:
+    #     os.makedirs(target_dir, exist_ok=True)
+    speech_path_list = get_path_list(speech_dir, "*.wav")
+    rir_path_list = get_path_list(rir_dir, "*.rir")
+
+    sentence_num = 20000
+    sr = 16000
+    volume_range = [0.5, 1.0]
+    sentence_len = sr * 10
+    # for i in range(10):
+    #     babble_noise_generate(i, speech_path_list, rir_path_list)
+
+    # 多核并行处理
+    # Parallel(n_jobs=-1)(delayed(babble_noise_generate)(i, speech_path_list, rir_path_list) for i in range(sentence_num))
+
+    pool = Pool(32)  # 创建进程池, 一定要放在 事件函数之后, 不然会保错
+
+    for i in range(sentence_num):
+        r = pool.apply_async(func=babble_noise_generate, args=(i, speech_path_list, rir_path_list,))
+
+    pool.close()  # 关闭进程池
+    pool.join()  # 回收进程池

neverlib-0.2.9/neverlib/tests/__init__.py

@@ -0,0 +1,16 @@
+# 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=[
+        "test_imports",
+        "test_preprocess",
+        "test_vad",
+    ],
+    submod_attrs={
+        "test_preprocess": ['test_NS_shape', 'test_pre_emphasis'],
+        "test_vad": ['test_vad2nad'],
+    }
+)

neverlib-0.2.9/neverlib/utils/README.md

@@ -0,0 +1,29 @@
+prsync
+来源：https://github.com/nathanhaigh/parallel-rsync
+用法
+如果您的命令如下所示：rsync
+
+```bash
+rsync \
+  --times --recursive --progress \
+  --exclude "raw_reads" --exclude ".snakemake" \
+  user@example.com:/my_remote_dir/ /my_local_dir/
+```
+只需替换此脚本的可执行文件：rsync
+```bash
+./prsync \
+  --times --recursive --progress \
+  --exclude "raw_reads" --exclude ".snakemake" \
+  user@example.com:/my_remote_dir/ /my_local_dir/
+  ```
+
+并行作业数
+默认情况下，脚本将为计算机上的每个处理器使用 1 个并行作业。 这是由确定的，如果失败，我们将回退到并行作业来传输文件。 
+可以通过使用作为脚本的第一个命令行参数来覆盖此行为：```nproc 10 --parallel```
+```bash
+./prsync \
+  --parallel=20 \
+  --times --recursive --progress \
+  --exclude "raw_reads" --exclude ".snakemake" \
+  user@example.com:/my_remote_dir/ /my_local_dir/
+```

neverlib-0.2.9/neverlib/utils/__init__.py

@@ -0,0 +1,27 @@
+# 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=[
+        "audio_split",
+        "checkGPU",
+        "floder",
+        "lazy_expose",
+        "lazy_module",
+        "message",
+        "pcm",
+        "utils",
+    ],
+    submod_attrs={
+        "audio_split": ['audio_split_VADfunasr', 'audio_split_VADsilero', 'audio_split_ffmpeg', 'audio_split_np', 'audio_split_pydub', 'audio_split_random', 'audio_split_sox'],
+        "checkGPU": ['get_gpu_utilization', 'get_gpu_utilization2', 'monitor_gpu_utilization'],
+        "floder": ['change_path', 'del_empty_folders', 'get_leaf_folders', 'rename_files_and_folders'],
+        "lazy_expose": ['attach_and_expose_all'],
+        "lazy_module": ['LazyModule'],
+        "message": ['send_QQEmail', 'send_QQEmail_with_images'],
+        "pcm": ['pcm2wav', 'read_pcm', 'wav2pcm'],
+        "utils": ['DatasetSubfloderSplit', 'TrainValSplit', 'TrainValTestSplit', 'get_audio_segments', 'get_file_time', 'get_path_list', 'save_weight_histogram'],
+    }
+)