spectrogram.py

import numpy as np
import matplotlib.pyplot as plt
from scipy.io import wavfile

def spectrogram(audio_file):
 rate, data = wavfile.read(audio_file)
 if len(data.shape) > 1:
 data = data.mean(axis=1)
 
 window_size = 1024
 hop_size = 512
 
 n_samples = len(data)
 n_overlap = window_size - hop_size
 n_windows = (n_samples - n_overlap) // hop_size
 
 spectrogram = np.zeros((n_windows, window_size // 2 + 1))
 
 for i in range(n_windows):
 start = i * hop_size
 end = start + window_size
 segment = data[start:end]
 windowed_segment = segment * np.hamming(window_size)
 spectrum = np.abs(np.fft.rfft(windowed_segment, window_size))
 spectrogram[i] = spectrum
 
 plt.figure(figsize=(7, 5))
 plt.imshow(np.log1p(spectrogram.T), aspect='auto', origin='lower')
 plt.colorbar(format='%+2.0f dB')
 plt.title('Spectrogram')
 plt.xlabel('Time')
 plt.ylabel('Frequency')
 plt.show()

audio_file_path = './data/file_example_WAV_2MG.wav'
spectrogram(audio_file_path)