img2wav_stft.py

import os
import sys
import cv2
import numpy as np
from scipy.io import wavfile
import matplotlib.pyplot as plt


def image_to_audio(image_path, output_wav, output_spectrogram):
    img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
    height, width = img.shape
    spectrogram = []
    for col in range(width):
        column_data = img[:, col]
        f_col = np.fft.fft(column_data)
        magnitude_col = 20 * np.log(np.abs(f_col))
        spectrogram.append(magnitude_col)
    spectrogram = np.array(spectrogram).T

    plt.figure(figsize=(10, 6))
    plt.imshow(spectrogram, cmap="jet", aspect="auto", origin="lower")
    plt.title("Spectrogram (Time on X-axis)")
    plt.xlabel("Time (Columns of Image)")
    plt.ylabel("Frequency (FFT Magnitude)")
    plt.colorbar()
    plt.savefig(output_spectrogram)
    plt.close()

    f = np.fft.fft2(img)
    fshift = np.fft.fftshift(f)
    
    magnitude_spectrum = 20 * np.log(np.abs(fshift))
    plt.figure(figsize=(6, 6))
    plt.imshow(magnitude_spectrum, cmap='gray')
    plt.title('Magnitude Spectrum')
    plt.colorbar()
    plt.savefig(output_spectrogram)
    plt.close()

    inverse_f = np.fft.ifftshift(fshift)
    img_reconstructed = np.fft.ifft2(inverse_f)

    audio_signal = np.real(img_reconstructed).flatten()
    audio_signal_normalized = np.int16(
        (audio_signal / np.max(np.abs(audio_signal))) * 32767
    )
    sample_rate = 44100
    wavfile.write(output_wav, sample_rate, audio_signal_normalized)

    print(f"Audio saved as {output_wav}")
    print(f"Time-based Spectrogram saved as {output_spectrogram}")


if __name__ == "__main__":
    if len(sys.argv) < 2:
        print("Usage: python script_name.py image_file")
        sys.exit(1)

    img_file = sys.argv[1]
    basename = os.path.basename(img_file)
    wav_file = basename + ".wav"
    spectro_file = "output_time_spectrogram.png"

    image_to_audio(img_file, wav_file, spectro_file)