bdc-divya's picture
Update app.py
2e90272
# Import necessary libraries
import tempfile # Library for temporary file operations
from audiocraft.models import MusicGen # Import the MusicGen model for music generation
from audiocraft.data.audio import audio_write # Import audio_write function for saving audio
import gradio as gr # Import Gradio for creating a web interface
import torch # Import PyTorch for deep learning operations
import uuid # Library for generating unique identifiers (UUIDs)
import os # Operating system-related functions
from scipy.io.wavfile import write # Function for writing audio to a WAV file
# Load the model and set parameters
model = MusicGen.get_pretrained("facebook/musicgen-small") # Load a pre-trained MusicGen model
model.set_generation_params(duration=5) # Set the duration for audio generation in seconds
# Function to generate music from text descriptions
def generate_music(description):
# Generate audio based on the provided description
wav = model.generate([description]) # Generate audio using the model
audio_array = wav.cpu().numpy().squeeze() # Convert the audio to a NumPy array
sample_rate = model.sample_rate # Get the audio sample rate (e.g., 44100 Hz)
# Generate a unique file identifier (UUID) for the temporary audio file
file_id = uuid.uuid1()
# Define the path for the temporary audio file using the identifier
file_path = os.path.join(
tempfile.gettempdir(), # Get the temporary directory path
f'{file_id}.wav' # Create a unique file name with the UUID and .wav extension
)
# Add debugging statements to check file paths and directory permissions
print(f"Temporary directory: {tempfile.gettempdir()}")
print(f"File path: {file_path}")
# Write the generated audio to the temporary file as a WAV file
write(file_path, rate=sample_rate, data=audio_array)
# Return the path to the temporary audio file
return file_path
# Define Gradio interface with temporary file output
iface = gr.Interface(
fn=generate_music, # Use the generate_music function for processing input
inputs="text", # Accept text input from the user
outputs=gr.components.Audio(type="filepath", label="Audio"), # Display the generated audio as output
title="Text to Audio Generation", # Set the title of the web interface
description="Generate audio based on text descriptions.", # Provide a description
live=False, # Set to False if you don't want real-time updates (for beginner-friendly interaction)
)
# Start the Gradio interface
iface.launch(debug=True)