Librosa无法绘制生成的mfcc

Question

我本来可以从系统捕获的音频中生成MFCC并绘制它，但经过一些重构和使用CUDA配置Tensorflow之后。我使用Librosa生成了mfcc，matplotlib.pyplot和Librosa.display来绘制mfcc和声音设备，从窗口的立体声混音中捕捉声音。当前配置可以从样例.wav文件创建和绘制MFCC，但当使用系统捕获的声音时，它无法绘制，因为它在运行MFCC时生成的是3D阵列而不是2D阵列。这是生成和绘制的代码

N_MFCC = 40 
N_MELS = 40 
N_FFT = 512 
HOP_LENGTH = 160 
MIN_FREQ = 0 
MAX_FREQ = None
def create_mfcc(record, sample_rate):
features = librosa.feature.mfcc(record, sample_rate, n_fft=N_FFT,n_mfcc=N_MFCC, 
n_mels=N_MELS,hop_length=HOP_LENGTH,fmin=MIN_FREQ, fmax=MAX_FREQ, htk=False)
return features
def plot_and_save_mfcc(mfcc_data, file_name, sample_rate):
plt.figure(figsize=(10, 8))
plt.title('Current audio MFCC', fontsize=18)
plt.xlabel('Time [s]', fontsize=18)
librosa_display.specshow(mfcc_data, sr=sample_rate)
plt.savefig(file_name)
plt.cla()

这将生成此堆栈跟踪

Traceback (most recent call last):
File "main.py", line 68, in <module>
main()
File "main.py", line 63, in main
start_listening_and_creating_mfcc()
File "main.py", line 48, in start_listening_and_creating_mfcc
plot_and_save_mfcc(mfcc_data, conf.DEFAULT_MFCC_IMAGE_NAME.format(image_count), conf.SAMPLE_RATE)
File "main.py", line 38, in plot_and_save_mfcc
librosa_display.specshow(mfcc_data, sr=sample_rate)
File anaconda3libsite-packageslibrosautildecorators.py", line 88, in inner_f
return f(*args, **kwargs)
File anaconda3libsite-packageslibrosadisplay.py", line 879, in specshow
out = axes.pcolormesh(x_coords, y_coords, data, **kwargs)
File anaconda3libsite-packagesmatplotlib__init__.py", line 1361, in inner
return func(ax, *map(sanitize_sequence, args), **kwargs)
File anaconda3libsite-packagesmatplotlibaxes_axes.py", line 6183, in pcolormesh
X, Y, C, shading = self._pcolorargs('pcolormesh', *args,
File anaconda3libsite-packagesmatplotlibaxes_axes.py", line 5671, in _pcolorargs
nrows, ncols = C.shape
ValueError: too many values to unpack (expected 2)

我确实试过调试它并更改mfcc配置，但没有成功。我也尝试重新配置我的环境，但这也没有帮助。

编辑：这是mfcc。系统音频的形状(48000，40，1(对于.wav示例文件(40122(

如前所述，我遗漏了一个函数，但这里是它和用于为.wav文件加载和创建mfcc的函数

def create_mfcc_from_file(file_path):
(signal, sample_rate) = librosa.load(file_path)
librosa_features = create_mfcc(signal, sample_rate)
plot_and_save_mfcc(librosa_features, 'mfcc-librosa', sample_rate)
def start_listening_and_creating_mfcc():
image_count = 0
while True:
my_recording = record_window()
mfcc_data = create_mfcc(my_recording, conf.SAMPLE_RATE)
plot_and_save_mfcc(mfcc_data, conf.DEFAULT_MFCC_IMAGE_NAME.format(image_count), conf.SAMPLE_RATE)
wav.write(conf.DEFAULT_MFCC_IMAGE_NAME.format(image_count) + '.wav', conf.SAMPLE_RATE, my_recording)
image_count += 1

Answer 1

def delta(feat, N):
"""Compute delta features from a feature vector sequence.
:param feat: A numpy array of size (NUMFRAMES by number of features) containing features. Each row holds 1 feature vector.
:param N: For each frame, calculate delta features based on preceding and following N frames
:returns: A numpy array of size (NUMFRAMES by number of features) containing delta features. Each row holds 1 delta feature vector.
"""
if N < 0:
raise ValueError('N must be an integer >0')
NUMFRAMES = len(feat)
denominator = 2 * sum([i**2 for i in range(1, N+1)])
delta_feat = numpy.empty_like(feat)
padded = numpy.pad(feat, ((N, N), (0, 0)), mode='edge')   # padded version of feat
for t in range(NUMFRAMES):

delta_feat[t] = numpy.dot(numpy.arange(-N, N+1), padded[t : t+2*N+1]) / denominator 
plt.plot(signal, c='c')# [t : t+2*N+1] == [(N+t)-N : (N+t)+N+1]
return delta_feat