image to sound

horn.py

from PIL import Image
from PIL import ImageFilter
import numpy as np
import matplotlib.pyplot as plt
from scipy.io.wavfile import write

threshold = 200

# Open image and convert to numpy array
img = Image.open("matterhorn.png")
# Apply Gaussian blur
# img = img.filter(ImageFilter.GaussianBlur(radius=1))

a = np.array(img)

mask = np.linalg.norm(a-a[0,0], ord=2, axis=2)
mask = np.where(mask>threshold, 255.0, 0.0)
mask = mask.cumsum(axis=0)
mask = mask.clip(0.0, 255.0)

# Convert numpy array back to image
new_img = Image.fromarray(mask)

# Save the new image
new_img.save("mask.gif")

window_size = 1000
sample_rate = 48000

contour = np.argmax(mask, axis=0)

contour = (-contour + np.max(contour))
contour = 10**(contour.astype(np.float64)/20)
angle = np.exp(1j * 2 * np.pi * np.random.rand(len(contour)))
peak = np.argmax(contour)
contour = angle * contour
steps = range(len(contour)-window_size+1)
for i in steps:
    step_peak = peak - i
    if step_peak < 0:
        break
    # print(step_peak)
    freq_bin = step_peak / (window_size * 2 + 1) * sample_rate
    # print(freq_bin)

    window = contour[i:i+window_size]
    window = np.hstack((0, window, window[::-1]))

    wave = np.real(np.fft.ifft(window))

    wave /= np.max(np.abs(wave))
    # wave = np.tile(wave, 10)
    write(f'sample/wave{int(freq_bin):05d}.wav', sample_rate, wave)

    # bounds = (i, 0, i+window_size, img.height)
    # win = img.crop(bounds)
    # win.save(f'img/win{int(freq_bin)}.png')
    # mwin = new_img.crop(bounds)
    # mwin.save(f'mask/mwin{int(freq_bin)}.gif')


# plt.plot(sound)
# plt.show()