Python语音信号处理：波形、频谱和声谱图可视化

import librosa
import matplotlib
import numpy as np
import matplotlib.pyplot as plt
from scipy.fft import fft
import librosa.display



plt.figure(dpi=600) # 将显示的所有图分辨率调高
matplotlib.rc("font",family='SimHei') # 显示中文
matplotlib.rcParams['axes.unicode_minus']=False # 显示符号


def displayWaveform(sample1, sample2): # 显示语音时域波形
    '''
    display waveform of a given speech sample
    :param sample_name: speech sample name
    :param fs: sample frequency
    :return:
    '''
    samples1, sr1 = librosa.load(sample1, sr=16000)
    samples2, sr2 = librosa.load(sample2, sr=16000)
    # samples = samples[6000:16000]

    print(len(samples1), sr1)
    print(len(samples2), sr2)
    time1 = np.arange(0, len(samples1)) * (1.0 / sr1)
    time2 = np.arange(0, len(samples2)) * (1.0 / sr2)

    plt.figure(figsize=(18, 8))
    plt.subplot(211)
    plt.plot(time1, samples1)
    plt.title("语音信号1时域波形")
    plt.xlabel("时长（秒）")
    plt.ylabel("振幅")

    plt.subplot(212)
    #plt.ylim(-0.2, 0.2)
    plt.plot(time2, samples2)
    plt.title("语音信号2时域波形")
    plt.xlabel("时长（秒）")
    plt.ylabel("振幅")
    plt.subplots_adjust(hspace=0.5) # 调整子图间距
    # plt.savefig("your dir\语音信号时域波形图", dpi=600)
    plt.show()

def displaySpectrum(sample1, sample2): # 显示语音频域谱线
    x1, sr1 = librosa.load(sample1, sr=16000)
    x2, sr2 = librosa.load(sample2, sr=16000)
    print(len(x1), len(x2))
    # ft = librosa.stft(x)
    # magnitude = np.abs(ft)  # 对fft的结果直接取模（取绝对值），得到幅度magnitude
    # frequency = np.angle(ft)  # (0, 16000, 121632)

    ft1 = fft(x1)
    ft2 = fft(x2)
    magnitude1 = np.absolute(ft1)  # 对fft的结果直接取模（取绝对值），得到幅度magnitude
    magnitude2 = np.absolute(ft2)  # 对fft的结果直接取模（取绝对值），得到幅度magnitude
    frequency1 = np.linspace(0, sr1, len(magnitude1))  # (0, 16000, 121632)
    frequency2 = np.linspace(0, sr2, len(magnitude2))  # (0, 16000, 121632)

    print(len(magnitude1), type(magnitude1), np.max(magnitude1), np.min(magnitude1))
    print(len(frequency1), type(frequency1), np.max(frequency1), np.min(frequency1))
    print(len(magnitude2), type(magnitude2), np.max(magnitude2), np.min(magnitude2))
    print(len(frequency2), type(frequency2), np.max(frequency2), np.min(frequency2))

    # plot spectrum，限定[:40000]
    plt.figure(figsize=(18, 8))
    plt.plot(frequency1[:40000], magnitude1[:40000], label='原始语音')  # magnitude spectrum
    plt.plot(frequency2[:40000], magnitude2[:40000], label='增强语音')  # magnitude spectrum
    plt.title("语音信号频域谱线")
    plt.xlabel("频率（赫兹）")
    plt.ylabel("幅度")
    plt.legend()
    # plt.savefig("your dir\语音信号频谱图", dpi=600)
    plt.show()


def displaySpectrogram(sample1, sample2):
    x1, sr1 = librosa.load(sample1, sr=16000)
    x2, sr2 = librosa.load(sample2, sr=16000)

    # compute power spectrogram with stft(short-time fourier transform):
    # 基于stft，计算power spectrogram
    spectrogram1 = librosa.amplitude_to_db(librosa.stft(x1))
    spectrogram2 = librosa.amplitude_to_db(librosa.stft(x2))

    # show
    plt.figure(figsize=(18, 8))
    plt.subplot(211)
    librosa.display.specshow(spectrogram1, y_axis='log')
    plt.colorbar(format='%+2.0f dB')
    plt.title('语音信号1对数谱图')
    plt.xlabel('时长（秒）')
    plt.ylabel('频率（赫兹）')
    plt.subplot(212)
    librosa.display.specshow(spectrogram2, y_axis='log')
    plt.colorbar(format='%+2.0f dB')
    plt.title('语音信号2对数谱图')
    plt.xlabel('时长（秒）')
    plt.ylabel('频率（赫兹）')
    plt.subplots_adjust(hspace=0.5) # 调整子图间距
    plt.show()


if __name__ == '__main__':
    sample1 = r'p376_295.wav'
    sample2 = r'enhanced_p376_295.wav'
    displayWaveform(sample1, sample2)
    displaySpectrum(sample1, sample2)
    displaySpectrogram(sample1, sample2)

让对比图中的横纵坐标保持一致，第二个图和第一个图一致内容：修改displaySpectrum函数中的以下两行代码：

plt.subplot(211)
plt.plot(frequency1[:40000], magnitude1[:40000])  # magnitude spectrum

修改为：

plt.figure(figsize=(18, 8))
plt.plot(frequency1[:40000], magnitude1[:40000], label='原始语音')  # magnitude spectrum
plt.plot(frequency2[:40000], magnitude2[:40000], label='增强语音')  # magnitude spectrum
plt.title("语音信号频域谱线")
plt.xlabel("频率（赫兹）")
plt.ylabel("幅度")
plt.legend()
# plt.savefig("your dir\语音信号频谱图", dpi=600)
plt.show()