How to plot Spectrogram using STFT in python?

Question

I calculated STFT of uint8 I/Q data and stored it in a numpy matrix where each row stores STFT of one window as shown in sudo code below.

#k= length of window
#fs= Sampling frequency
#n= Number of STFT calculated
#matrix= Initially empty numpy array

for i in range(0,n):
  t=data[start:end,:]   #start & end calculated with each iteration
  t=t.flatten()
  t=t-127.5
  array = np.empty(t.shape[0]//2, dtype=np.complex128)
  array.real = t[::2]
  array.imag = t[1::2]

  transform=(np.fft.fft(temp_array))
  line = 2*abs(transform)/k

  #Inserting row into numpy array
  if(i==0):
     matrix = np.hstack((matrix, line))
  else:
     matrix = np.vstack((matrix, line))

Now how can I plot frequency vs time Spectrogram ?

Solution 1

what you can do is use this

first method:

import scipy.io.wavfile as wav
import scipy.signal as signal
from matplotlib import pyplot as plt

sample_rate, samples = wav.read(filename)
f, t, Zxx = signal.stft(samples, fs=sample_rate)
plt.pcolormesh(t, f, np.abs(Zxx), cmap=cmap)

or the second method:

plt.specgram(samples, cmap=cmap, Fs=sample_rate)

I found the second one have a better visualization effect. Not sure how to make the first one look as good as the second one though.

Solution 2

import os
import librosa
import librosa.display
import IPython.display as ipd
import numpy as np
import matplotlib.pyplot as plt

filename = '<yourfile name/ location>'
x, sr = librosa.load(filename)

import librosa.display                   #explicitly import librosa.display
X = librosa.stft(x)                      #perform short-term fourier transfrom               
Xdb = librosa.amplitude_to_db(abs(X))    #convert an amplitude spectrogram to dB-scaled spectrogram.
plt.figure(figsize=(15, 5))              #initialize the fig size
librosa.display.specshow(Xdb, sr=sr, x_axis='time', y_axis='hz')
plt.colorbar()```

Comments

[user2027202827]

have you looked into matplotlib?

[Jay Krishna]

Yes I did both into matplotlib as well as scipy but both were not working fine with complex numpy array

[MB-F]

Why complex arrays? Taking the absolute value of the FFT should result in real values that can be easily plotted with matplotlib's imshow. Just take care not to initialize matrix with complex data type.

[Jay Krishna]

@kazemakase You mean using imshow with extent after taking absolute i.e ax.imshow(np.absolute(matrix),extent=[0,100,0,1]) ? I am assuming frequency will be on Y-axis.

[MB-F]

@JayKrishna yes, that should work as expected. Stil... it looks like you only put absolute values into line and thus in matrix anyway so it is surprising that you have complex values in matrix.

[Ahmed Fasih]

See the example at the bottom of docs.scipy.org/doc/scipy/reference/generated/…

[Tom Wyllie]

Possible duplicate of How to convert a .wav file to a spectrogram in python3

[Chase Roberts]

cmap is undefined.

[jcoppens]

(second) cmap must be eplaced with the name of a colormap. In the Gallery on matplotlib.org is a drawing with all available colormaps, their names. and how to use them.

[jcoppens]

Oops. Gallery is not on the front page anymore. It's under examples now. Check this: matplotlib.org/gallery/color/…

[alienflow]

To achieve on the first the looks of second just do np.log() of np.abs(Zxx)