app.py

import gradio as gr
import torchaudio
import tempfile
from audiocraft.models import MusicGen
from bark import SAMPLE_RATE as BARK_SAMPLE_RATE, generate_audio as bark_generate_audio
from pydub import AudioSegment

# Load MusicGen model
musicgen = MusicGen.get_pretrained('facebook/musicgen-small')

# Function to generate vocals and music
def generate_song(lyrics, genre_prompt):
    # Step 1: Generate vocals from Bark
    vocals = bark_generate_audio(lyrics, history_prompt="v2/en_speaker_6")
    
    # Save Bark vocals temporarily
    vocals_path = tempfile.mktemp(suffix=".wav")
    torchaudio.save(vocals_path, vocals.squeeze(0).cpu(), BARK_SAMPLE_RATE)

    # Step 2: Generate music from MusicGen
    musicgen.set_generation_params(duration=15)  # Set song length (in seconds)
    music = musicgen.generate([genre_prompt])  # Generate instrumental based on genre
    music_path = tempfile.mktemp(suffix=".wav")
    torchaudio.save(music_path, music[0].cpu(), 32000)

    # Step 3: Mix the two audio files using pydub
    vocals_seg = AudioSegment.from_wav(vocals_path)
    music_seg = AudioSegment.from_wav(music_path)

    # Overlay vocals over the music
    mixed = music_seg.overlay(vocals_seg.set_frame_rate(32000).set_channels(1))
    output_path = tempfile.mktemp(suffix=".wav")
    mixed.export(output_path, format="wav")

    return output_path

# Gradio interface
iface = gr.Interface(
    fn=generate_song,
    inputs=[
        gr.Textbox(label="Enter Lyrics", lines=4),
        gr.Textbox(label="Enter Genre (e.g., 'hip-hop with 808s')")
    ],
    outputs=gr.Audio(label="Generated Song")
)

# Launch the Gradio interface
iface.
launch()