认识提取与显示梅尔谱图的小实验（观察不同y_axis和x_axis的区别）

导入librosa

import librosa

读取音频

y, sr = librosa.load("C:/Users/24061/Desktop/MER数据集/DEAM/DEAM_audio/MEMD_audio_wav/2.wav")

y

array([0.        , 0.        , 0.        , ..., 0.4163208 , 0.43338013,
       0.40551758], dtype=float32)

sr

22050

提取梅尔谱图

mel_spectrogram = librosa.feature.melspectrogram(y=y, sr=sr, n_fft=2048, hop_length=1024)

mel_spectrogram  # type: numpy.ndarray

array([[0.00000000e+00, 1.25761675e-02, 2.62945890e+00, ...,
        4.08293676e+00, 7.79196739e+00, 5.92219353e+00],
       [0.00000000e+00, 1.00206733e-01, 1.33076525e+00, ...,
        6.49990678e-01, 1.44000304e+00, 1.67580545e+00],
       [0.00000000e+00, 4.64823037e-01, 1.54586525e+01, ...,
        3.54503012e+00, 2.53890848e+00, 9.59981441e+00],
       ...,
       [0.00000000e+00, 6.95451519e-09, 3.43443826e-05, ...,
        6.05733460e-03, 1.72329806e-02, 7.06060929e-03],
       [0.00000000e+00, 7.65795605e-09, 7.63881962e-06, ...,
        1.81941327e-03, 3.55470460e-03, 4.70093498e-03],
       [0.00000000e+00, 4.74458783e-09, 5.26388646e-07, ...,
        1.27859021e-04, 7.03962069e-05, 1.91266462e-03]], dtype=float32)

mel_spectrogram.shape

(128, 971)

显示梅尔谱图

1、没有转化为对数谱时

• 以梅尔标度显示频率y_axis='mel'

import librosa.display
librosa.display.specshow(mel_spectrogram, y_axis='mel', x_axis='time')

<matplotlib.collections.QuadMesh at 0x23d902a60b8>

• 以对数标度显示频率y_axis='log'

librosa.display.specshow(mel_spectrogram, y_axis='log', x_axis='time')

<matplotlib.collections.QuadMesh at 0x23d92827e80>

2、将幅度转化为对数

mel_spectrogram_db = librosa.amplitude_to_db(mel_spectrogram)

mel_spectrogram_db

array([[-19.654686 , -19.654686 ,   8.397327 , ...,  12.219453 ,
         17.832943 ,  15.449652 ],
       [-19.654686 , -19.654686 ,   2.482029 , ...,  -3.7418575,
          3.167268 ,   4.484472 ],
       [-19.654686 ,  -6.6542473,  23.783434 , ...,  10.992398 ,
          8.092941 ,  19.645256 ],
       ...,
       [-19.654686 , -19.654686 , -19.654686 , ..., -19.654686 ,
        -19.654686 , -19.654686 ],
       [-19.654686 , -19.654686 , -19.654686 , ..., -19.654686 ,
        -19.654686 , -19.654686 ],
       [-19.654686 , -19.654686 , -19.654686 , ..., -19.654686 ,
        -19.654686 , -19.654686 ]], dtype=float32)

mel_spectrogram_db.shape

(128, 971)

• 以梅尔标度显示频率y_axis="mel"

librosa.display.specshow(mel_spectrogram_db, y_axis="mel",x_axis="time" )

<matplotlib.collections.QuadMesh at 0x23d9194ce10>

• 以对数标度显示频率y_axis="log"

librosa.display.specshow(mel_spectrogram_db, y_axis="log",x_axis="time" )

<matplotlib.collections.QuadMesh at 0x23d925c7cc0>

以Hz显示频率y_axis="hz"

librosa.display.specshow(mel_spectrogram_db, y_axis="hz",x_axis="time" )

<matplotlib.collections.QuadMesh at 0x23d92c2f5f8>

• 以对数谱显示频率y_axis="log"，时间单位为秒x_axis="s"

librosa.display.specshow(mel_spectrogram_db, y_axis="log",x_axis="s" )

<matplotlib.collections.QuadMesh at 0x23d9348a550>

• 以对数谱显示频率y_axis="log"，时间单位为毫秒x_axis="ms"

librosa.display.specshow(mel_spectrogram_db, y_axis="log",x_axis="ms" )

<matplotlib.collections.QuadMesh at 0x23d93ea76a0>

---

节奏（待续），还没搞太明白

librosa.feature.fourier_tempogram(y, sr)

C:\Users\24061\anaconda3\envs\tensorflow\lib\site-packages\ipykernel_launcher.py:1: FutureWarning: Pass y=[0.         0.         0.         ... 0.4163208  0.43338013 0.40551758], sr=22050 as keyword args. From version 0.10 passing these as positional arguments will result in an error
  """Entry point for launching an IPython kernel.





array([[ 1.41953934e+02+0.0000000e+00j,  1.43232498e+02+0.0000000e+00j,
         1.44507858e+02+0.0000000e+00j, ...,
         1.20833031e+02+0.0000000e+00j,  1.19599785e+02+0.0000000e+00j,
         1.18365807e+02+0.0000000e+00j],
       [-8.12093430e+01+7.8693253e+01j, -8.25004044e+01+7.8347717e+01j,
        -8.37830200e+01+7.7980965e+01j, ...,
        -6.11498871e+01-7.6005348e+01j, -5.98955841e+01-7.6018913e+01j,
        -5.86419067e+01-7.6011627e+01j],
       [ 2.08344612e+01-5.4645943e+01j,  2.22085571e+01-5.3934937e+01j,
         2.35550823e+01-5.3178368e+01j, ...,
         1.26519930e+00+5.0331814e+01j,  4.82287928e-02+5.0350494e+01j,
        -1.16933417e+00+5.0330265e+01j],
       ...,
       [-3.68897580e-02-7.4101496e-01j,  9.78471190e-02+7.0739186e-01j,
        -1.57261893e-01-6.6958255e-01j, ...,
        -4.85646218e-01+1.2208136e-01j,  4.81895536e-01-1.4795184e-01j,
        -4.76868808e-01+1.7341925e-01j],
       [-1.62224078e+00-6.7166932e-02j,  1.64763165e+00+3.6856860e-02j,
        -1.67323339e+00-5.3153611e-03j, ...,
         6.40185595e-01-6.1752874e-01j, -6.37307167e-01+6.1541033e-01j,
         6.34444416e-01-6.1347997e-01j],
       [ 1.42917812e+00+0.0000000e+00j, -1.40549254e+00+0.0000000e+00j,
         1.38058436e+00+0.0000000e+00j, ...,
        -1.41562808e+00+0.0000000e+00j,  1.39833307e+00+0.0000000e+00j,
        -1.38108861e+00+0.0000000e+00j]], dtype=complex64)

librosa.display.specshow(librosa.amplitude_to_db(librosa.feature.fourier_tempogram(y, sr)))

C:\Users\24061\anaconda3\envs\tensorflow\lib\site-packages\ipykernel_launcher.py:1: FutureWarning: Pass y=[0.         0.         0.         ... 0.4163208  0.43338013 0.40551758], sr=22050 as keyword args. From version 0.10 passing these as positional arguments will result in an error
  """Entry point for launching an IPython kernel.
C:\Users\24061\anaconda3\envs\tensorflow\lib\site-packages\librosa\util\decorators.py:88: UserWarning: amplitude_to_db was called on complex input so phase information will be discarded. To suppress this warning, call amplitude_to_db(np.abs(S)) instead.
  return f(*args, **kwargs)





<matplotlib.collections.QuadMesh at 0x23d9325f860>