inference.py

import os
import numpy as np
import matplotlib.pyplot as plt
from scipy.io import wavfile
from scipy import signal
from PIL import Image

def create_spectrogram(wav_file, output_folder):
    # Read the wav file
    sample_rate, data = wavfile.read(wav_file)
    
    # Convert data to mono if it's stereo
    if data.ndim == 2:
        data = np.mean(data, axis=1)
    
    # Create the spectrogram
    frequencies, times, spectrogram = signal.spectrogram(data, sample_rate)
    
    # Create the figure
    plt.figure(figsize=(10, 4))
    
    # Plot the spectrogram
    plt.pcolormesh(times, frequencies, 10 * np.log10(spectrogram), shading='gouraud', cmap='inferno')
    
    # Set the y-axis limit to the Nyquist frequency
    plt.ylim(0, sample_rate / 2)

    # Remove axes and labels
    plt.axis('off')
    
    # Get the current axis
    ax = plt.gca()
    
    # Set the x-axis limits to start from 0 to the last time point
    ax.set_xlim(0, times[-1])
    
    # Fill the area to the right of the spectrogram with black
    ax.add_patch(plt.Rectangle((times[-1], 0), 10, sample_rate / 2, facecolor='black', edgecolor='none'))

    # Save the spectrogram image
    filename = os.path.splitext(os.path.basename(wav_file))[0] + '.png'
    plt.savefig(os.path.join(output_folder, filename), bbox_inches='tight', pad_inches=0)
    plt.close()

    # Convert white pixels to black
    convert_white_to_black(os.path.join(output_folder, filename))

def convert_white_to_black(image_path):
    # Open the image
    img = Image.open(image_path)
    
    # Convert the image to RGB (if not already in that mode)
    img = img.convert("RGB")
    
    # Get the data of the image
    data = np.array(img)

    # Create a mask for white pixels
    white_pixels = (data[:, :, 0] == 255) & (data[:, :, 1] == 255) & (data[:, :, 2] == 255)

    # Change white pixels to black
    data[white_pixels] = [0, 0, 0]

    # Create a new image from the modified data
    new_img = Image.fromarray(data)
    
    # Save the modified image
    new_img.save(image_path)

def convert_wav_to_spectrograms(input_folder, output_folder):
    # Create output folder if it doesn't exist
    if not os.path.exists(output_folder):
        os.makedirs(output_folder)

    # Iterate through all files in the input folder
    for file in os.listdir(input_folder):
        if file.endswith('.wav'):
            wav_file_path = os.path.join(input_folder, file)
            create_spectrogram(wav_file_path, output_folder)
            print(f"Converted {file} to spectrogram.")

if __name__ == "__main__":
    input_folder = 'dataset'  # Input folder containing WAV files
    output_folder = 'spectrograms'  # Output folder for spectrogram images
    convert_wav_to_spectrograms(input_folder, output_folder)