音频干扰检测（时域方法）

请注意注释掉的代码：逐个包络比对就不能加窗了。

import librosa

import numpy as np

from scipy.signal import windows

import matplotlib.pyplot as plt

# 读取音频文件

audio_file = 'sine.wav'

signal, sample_rate = librosa.load(audio_file, sr=None, mono=False)

# 检查通道数并处理信号

if signal.ndim > 1:

    num_channels = signal.shape[0]

    print(f"音频文件有 {num_channels} 个通道")

    # 如果是4通道,取第X个通道进行处理，这里示例取第4个通道（索引为3）

    if num_channels == 2:

        signal = signal[0, :]

else:

    # 如果信号是单通道，直接使用

    print("音频文件是单通道")

# 计算每个周期的采样点数

cycle_samples = int(sample_rate / 1000)

# # 创建汉宁窗

# window_length = cycle_samples * 1  # 窗长度为10个周期

# window = windows.hann(window_length)

# # 对信号的开头和结尾分别应用汉宁窗

# windowed_signal = signal.copy()

# windowed_signal[:window_length//2] *= window[:window_length//2]

# windowed_signal[-window_length//2:] *= window[window_length//2:]

# 计算周期数

num_cycles = len(signal) // cycle_samples

# 存储异常周期的时间点和幅值

anomaly_times = []

anomaly_amplitudes = []

# 逐个周期比较包络

for i in range(num_cycles - 1):

    start = i * cycle_samples

    end = (i + 1) * cycle_samples

    current_cycle = signal[start:end]

    next_cycle = signal[end:end+cycle_samples]

   

    # 计算当前周期和下一个周期的包络差异

    diff = np.abs(current_cycle - next_cycle)

   

    # 如果差异大于阈值,则认为是异常周期

    if np.max(diff) > 0.1:

        anomaly_time = start / sample_rate

        anomaly_times.append(anomaly_time)

        anomaly_amplitudes.append(np.max(np.abs(current_cycle)))

# 打印异常周期的时间点和幅值

for time, amplitude in zip(anomaly_times, anomaly_amplitudes):

    print(f"异常周期时间点: {time:.3f}s, 幅值: {amplitude:.3f}")

# 绘制时域波形图

time = np.arange(len(signal)) / sample_rate

plt.figure(figsize=(8,4))

plt.plot(time, signal, label='Signal')

# 标注异常周期

for t in anomaly_times:

    plt.axvline(x=t, color='r', linestyle='--', label='Anomaly Detected')

plt.xlabel('Time(s)')

plt.ylabel('Amplitude')

plt.title('Windowed Waveform with Anomalies Highlighted')

plt.legend()

plt.show()