app.py

import gradio as gr
import matplotlib.pyplot as plt
import librosa
import numpy as np
from PIL import Image, ImageDraw, ImageFont
from moviepy.editor import *
from moviepy.video.io.VideoFileClip import VideoFileClip

def make_bars_image(height_values, index, new_height):
    
    # Define the size of the image
    width = 512  
    height = new_height
    
    # Create a new image with a transparent background
    image = Image.new('RGBA', (width, height), color=(0, 0, 0, 0))
    
    # Get the image drawing context
    draw = ImageDraw.Draw(image)
    
    # Define the rectangle width and spacing
    rect_width = 2
    spacing = 2
    
    # Define the list of height values for the rectangles
    #height_values = [20, 40, 60, 80, 100, 80, 60, 40]
    num_bars = len(height_values)
    # Calculate the total width of the rectangles and the spacing
    total_width = num_bars * rect_width + (num_bars - 1) * spacing
    
    # Calculate the starting position for the first rectangle
    start_x = int((width - total_width) / 2)
    # Define the buffer size
    buffer_size = 80
    # Draw the rectangles from left to right
    x = start_x
    for i, height in enumerate(height_values):
        
        # Define the rectangle coordinates
        y0 = buffer_size
        y1 = height + buffer_size
        x0 = x
        x1 = x + rect_width

        # Draw the rectangle
        draw.rectangle([x0, y0, x1, y1], fill='white')  
        
        # Move to the next rectangle position
        if i < num_bars - 1:
            x += rect_width + spacing
        

    # Rotate the image by 180 degrees
    image = image.rotate(180)
    
    # Mirror the image
    image = image.transpose(Image.FLIP_LEFT_RIGHT)
    
    # Save the image
    image.save('audio_bars_'+ str(index) + '.png')

    return 'audio_bars_'+ str(index) + '.png'

def db_to_height(db_value):
    # Scale the dB value to a range between 0 and 1
    scaled_value = (db_value + 80) / 80
    
    # Convert the scaled value to a height between 0 and 100
    height = scaled_value * 50
    
    return height

def infer(title, audio_in, image_in, output_video_path):
    # Load the audio file
    audio_path = audio_in
    audio_data, sr = librosa.load(audio_path)

    # Get the duration in seconds
    duration = librosa.get_duration(y=audio_data, sr=sr)
    
    # Extract the audio data for the desired time
    start_time = 0 # start time in seconds
    end_time = duration # end time in seconds
    
    start_index = int(start_time * sr)
    end_index = int(end_time * sr)
    
    audio_data = audio_data[start_index:end_index]
    
    # Compute the short-time Fourier transform
    hop_length = 512

    
    stft = librosa.stft(audio_data, hop_length=hop_length)
    spectrogram = librosa.amplitude_to_db(np.abs(stft), ref=np.max)

    # Get the frequency values
    freqs = librosa.fft_frequencies(sr=sr, n_fft=stft.shape[0])

    # Select the indices of the frequency values that correspond to the desired frequencies
    n_freqs = 114
    freq_indices = np.linspace(0, len(freqs) - 1, n_freqs, dtype=int)
    
    # Extract the dB values for the desired frequencies
    db_values = []
    for i in range(spectrogram.shape[1]):
        db_values.append(list(zip(freqs[freq_indices], spectrogram[freq_indices, i])))
    
    # Print the dB values for the first time frame
    print(db_values[0])

    proportional_values = []

    for frame in db_values:
        proportional_frame = [db_to_height(db) for f, db in frame]
        proportional_values.append(proportional_frame)

    print(proportional_values[0])
    print("AUDIO CHUNK: " + str(len(proportional_values)))

    # Open the background image
    background_image = Image.open(image_in)
    
    # Resize the image while keeping its aspect ratio
    bg_width, bg_height = background_image.size
    aspect_ratio = bg_width / bg_height
    new_width = 512
    new_height = int(new_width / aspect_ratio)
    resized_bg = background_image.resize((new_width, new_height))

    # Apply black cache for better visibility of the white text
    bg_cache = Image.open('black_cache.png')
    resized_bg.paste(bg_cache, (0, resized_bg.height - bg_cache.height), mask=bg_cache)

    # Create a new ImageDraw object
    draw = ImageDraw.Draw(resized_bg)
    
    # Define the text to be added
    text = title
    font = ImageFont.truetype("Lato-Regular.ttf", 16)
    text_color = (255, 255, 255) # white color
    
    # Calculate the position of the text
    #text_width, text_height = draw.textsize(text, font=font)
    x = 30
    y = new_height - 70
    
    # Draw the text on the image
    draw.text((x, y), text, fill=text_color, font=font)

    # Save the resized image
    resized_bg.save('resized_background.jpg')
    
    generated_frames = []
    for i, frame in enumerate(proportional_values): 
        bars_img = make_bars_image(frame, i, new_height)
        bars_img = Image.open(bars_img)
        # Paste the audio bars image on top of the background image
        fresh_bg = Image.open('resized_background.jpg')
        fresh_bg.paste(bars_img, (0, 0), mask=bars_img)
        # Save the image
        fresh_bg.save('audio_bars_with_bg' + str(i) + '.jpg')
        generated_frames.append('audio_bars_with_bg' + str(i) + '.jpg')
    print(generated_frames)

    # Create a video clip from the images
    clip = ImageSequenceClip(generated_frames, fps=len(generated_frames)/(end_time-start_time))
    audio_clip = AudioFileClip(audio_in)
    clip = clip.set_audio(audio_clip)
    # Set the output codec
    codec = 'libx264'
    audio_codec = 'aac'
    # Save the video to a file
    clip.write_videofile("my_video.mp4", codec=codec, audio_codec=audio_codec)

    retimed_clip = VideoFileClip("my_video.mp4")

    # Set the desired frame rate
    new_fps = 25
    
    # Create a new clip with the new frame rate
    new_clip = retimed_clip.set_fps(new_fps)
    
    # Save the new clip as a new video file
    new_clip.write_videofile(output_video_path, codec=codec, audio_codec=audio_codec)
    
    # Visualize the audio bars
    plt.figure(figsize=(10, 4))
    librosa.display.specshow(spectrogram, sr=sr, x_axis='time', y_axis='log')
    plt.colorbar(format='%+2.0f dB')
    plt.title('Audio Bars Visualization')
    
    # Save the image as a JPG file
    output_path = 'image_out.jpg'
    plt.savefig(output_path, dpi=300, bbox_inches='tight')

    #test make image bars
    #bars_img = make_bars_image(proportional_values[0])
    return output_video_path, 'image_out.jpg'

gr.Interface(fn=infer, 
             inputs=[gr.Textbox(placeholder='FIND A GOOD TITLE'), 
                     gr.Audio(source='upload', type='filepath'), 
                     gr.Image(source='upload', type='filepath'),
                     gr.Textbox(label="Output video path", value="my_final_video.mp4", visible=False)], 
             outputs=[gr.Video(label='video result'), gr.Image(label='spectrogram image')],
            title='Animated Audio Visualizer', description='<p style="text-align: center;">Upload an audio file, upload a background image, choose a good title, click submit.</p>').launch()