app.py

import gradio as gr
import numpy as np
import librosa
import soundfile as sf
import pandas as pd

def generate_audio(clip_length=4.0, fade_in_duration=0.5, fade_out_duration=0.5, volume_factor=0.3):
    # Load audio files
    narration, sr = librosa.load('narration.wav', sr=None)
    baa, _ = librosa.load('baa.wav', sr=sr)
    murmur, _ = librosa.load('murmur.wav', sr=sr)
    
    # Calculate RMS for normalization
    narration_rms = np.sqrt(np.mean(narration**2))
    baa_rms = np.sqrt(np.mean(baa**2))
    murmur_rms = np.sqrt(np.mean(murmur**2))
    
    # Normalize baa and murmur to match narration volume
    baa_normalized = baa * (narration_rms / baa_rms)
    murmur_normalized = murmur * (narration_rms / murmur_rms)
    
    # Create output array with narration length
    output_length = len(narration)
    combined = np.zeros(output_length)
    
    # Add narration as baseline
    combined += narration
    
    # Add first clip_length seconds of baa at 0:05 (5 seconds) with fade in/out
    baa_clip = baa_normalized[:int(clip_length*sr)] * volume_factor
    
    # Create fade-in and fade-out envelopes
    # fade_in_duration and fade_out_duration are factors (0-1) of the clip length
    fade_in_samples = int(fade_in_duration * len(baa_clip))
    fade_out_samples = int(fade_out_duration * len(baa_clip))
    fade_in = np.linspace(0, 1, fade_in_samples)
    fade_out = np.linspace(1, 0, fade_out_samples)
    
    # Apply fade effects
    if fade_in_samples > 0:
        baa_clip[:fade_in_samples] *= fade_in
    if fade_out_samples > 0:
        baa_clip[-fade_out_samples:] *= fade_out
    
    start_idx = int(5 * sr)
    end_idx = start_idx + len(baa_clip)
    if end_idx <= output_length:
        combined[start_idx:end_idx] += baa_clip
    
    # Add first clip_length seconds of murmur at 0:15 (15 seconds) with fade in/out
    murmur_clip = murmur_normalized[:int(clip_length*sr)] * volume_factor
    
    # Calculate fade samples for murmur clip
    murmur_fade_in_samples = int(fade_in_duration * len(murmur_clip))
    murmur_fade_out_samples = int(fade_out_duration * len(murmur_clip))
    
    # Apply fade effects to murmur
    if murmur_fade_in_samples > 0:
        murmur_fade_in = np.linspace(0, 1, murmur_fade_in_samples)
        murmur_clip[:murmur_fade_in_samples] *= murmur_fade_in
    if murmur_fade_out_samples > 0:
        murmur_fade_out = np.linspace(1, 0, murmur_fade_out_samples)
        murmur_clip[-murmur_fade_out_samples:] *= murmur_fade_out
    
    start_idx = int(15 * sr)
    end_idx = start_idx + len(murmur_clip)
    if end_idx <= output_length:
        combined[start_idx:end_idx] += murmur_clip
    
    # Normalize to prevent clipping
    max_val = np.max(np.abs(combined))
    if max_val > 1.0:
        combined = combined / max_val
    
    return (sr, combined)

def visualize_sfx(sound_effect_clip_length, fade_in_duration, fade_out_duration, sound_effect_volume_factor):
    # Calculate fade durations in seconds
    fade_in_seconds = fade_in_duration * sound_effect_clip_length
    fade_out_seconds = fade_out_duration * sound_effect_clip_length
    
    # Create time array with high resolution for smooth visualization
    time_resolution = 0.01  # 10ms resolution
    times = np.arange(0, sound_effect_clip_length + time_resolution, time_resolution)
    
    # Calculate volume envelope
    volumes = []
    for t in times:
        if t <= fade_in_seconds and fade_in_seconds > 0:
            # Fade in phase
            volume = sound_effect_volume_factor * (t / fade_in_seconds)
        elif t >= sound_effect_clip_length - fade_out_seconds and fade_out_seconds > 0:
            # Fade out phase
            fade_out_progress = (sound_effect_clip_length - t) / fade_out_seconds
            volume = sound_effect_volume_factor * fade_out_progress
        else:
            # Steady state phase
            volume = sound_effect_volume_factor
        
        volumes.append(volume)
    
    # Create DataFrame for LinePlot
    plot_data = pd.DataFrame({
        "time": times,
        "volume": volumes
    })
    
    return plot_data

with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            sound_effect_clip_length = gr.Slider(minimum=0.5, maximum=5, value=4.0, step=0.1, label="Sound Effect Clip Length (seconds)")
            fade_in_duration = gr.Slider(minimum=0.0, maximum=1.0, value=0.2, step=0.05, label="Fade In Duration Factor", info="0.0 = no fade in, 1.0 = fade in over entire clip")
            fade_out_duration = gr.Slider(minimum=0.0, maximum=1.0, value=0.2, step=0.05, label="Fade Out Duration Factor", info="0.0 = no fade out, 1.0 = fade out over entire clip")
            sound_effect_volume_factor = gr.Slider(minimum=0.1, maximum=1.0, value=0.15, step=0.05, label="Sound Effect Volume Factor", info="0.1 is 10% of the narration volume, 1.0 is 100% of the original volume")
            visualization = gr.LinePlot(label="Sound Effect Volume Envelope", x="time", y="volume", y_lim=[0, 1])
            generate_button = gr.Button("Generate Audio")
        with gr.Column():
            output = gr.Audio()

    gr.on(
        [demo.load, sound_effect_clip_length.change, fade_in_duration.change, fade_out_duration.change, sound_effect_volume_factor.change],
        fn=visualize_sfx,
        inputs=[sound_effect_clip_length, fade_in_duration, fade_out_duration, sound_effect_volume_factor], 
        outputs=visualization
    )
    generate_button.click(generate_audio, inputs=[sound_effect_clip_length, fade_in_duration, fade_out_duration, sound_effect_volume_factor], outputs=output)

if __name__ == "__main__":
    demo.launch()