app.py

import gradio as gr
import subprocess
import os 
import re
import shutil
import tempfile

is_shared_ui = True if "fffiloni/YuE" in os.environ['SPACE_ID'] else False

# Install required package
def install_flash_attn():
    try:
        print("Installing flash-attn...")
        subprocess.run(
            ["pip", "install", "flash-attn", "--no-build-isolation"], 
            check=True
        )
        print("flash-attn installed successfully!")
    except subprocess.CalledProcessError as e:
        print(f"Failed to install flash-attn: {e}")
        exit(1)

# Install flash-attn
install_flash_attn()

from huggingface_hub import snapshot_download 

# Create xcodec_mini_infer folder
folder_path = './inference/xcodec_mini_infer'

# Create the folder if it doesn't exist
if not os.path.exists(folder_path):
    os.mkdir(folder_path)
    print(f"Folder created at: {folder_path}")
else:
    print(f"Folder already exists at: {folder_path}")

snapshot_download(
    repo_id = "m-a-p/xcodec_mini_infer",
    local_dir = "./inference/xcodec_mini_infer"
)

# Change to the "inference" directory
inference_dir = "./inference"
try:
    os.chdir(inference_dir)
    print(f"Changed working directory to: {os.getcwd()}")
except FileNotFoundError:
    print(f"Directory not found: {inference_dir}")
    exit(1)

def empty_output_folder(output_dir):
    # List all files in the output directory
    files = os.listdir(output_dir)
    
    # Iterate over the files and remove them
    for file in files:
        file_path = os.path.join(output_dir, file)
        try:
            if os.path.isdir(file_path):
                # If it's a directory, remove it recursively
                shutil.rmtree(file_path)
            else:
                # If it's a file, delete it
                os.remove(file_path)
        except Exception as e:
            print(f"Error deleting file {file_path}: {e}")

# Function to create a temporary file with string content
def create_temp_file(content, prefix, suffix=".txt"):
    temp_file = tempfile.NamedTemporaryFile(delete=False, mode="w", prefix=prefix, suffix=suffix)
    content = content.strip() + "\n\n"  # Add extra newline at end
    content = content.replace("\r\n", "\n").replace("\r", "\n")
    temp_file.write(content)
    temp_file.close()
    
    # Debug: Print file contents
    print(f"\nContent written to {prefix}{suffix}:")
    print(content)
    print("---")
    
    return temp_file.name

def get_last_mp3_file(output_dir):
    # List all files in the output directory
    files = os.listdir(output_dir)
    
    # Filter only .mp3 files
    mp3_files = [file for file in files if file.endswith('.mp3')]
    
    if not mp3_files:
        print("No .mp3 files found in the output folder.")
        return None
    
    # Get the full path for the mp3 files
    mp3_files_with_path = [os.path.join(output_dir, file) for file in mp3_files]
    
    # Sort the files based on the modification time (most recent first)
    mp3_files_with_path.sort(key=lambda x: os.path.getmtime(x), reverse=True)
    
    # Return the most recent .mp3 file
    return mp3_files_with_path[0]

def infer(genre_txt_content, lyrics_txt_content, num_segments, max_new_tokens):
    # Create temporary files
    genre_txt_path = create_temp_file(genre_txt_content, prefix="genre_")
    lyrics_txt_path = create_temp_file(lyrics_txt_content, prefix="lyrics_")

    print(f"Genre TXT path: {genre_txt_path}")
    print(f"Lyrics TXT path: {lyrics_txt_path}")

    # Ensure the output folder exists
    output_dir = "./output"
    os.makedirs(output_dir, exist_ok=True)
    print(f"Output folder ensured at: {output_dir}")

    empty_output_folder(output_dir)
 
    # Command and arguments with optimized settings
    command = [
        "python", "infer.py",
        "--stage1_model", "m-a-p/YuE-s1-7B-anneal-en-cot",
        "--stage2_model", "m-a-p/YuE-s2-1B-general",
        "--genre_txt", f"{genre_txt_path}",
        "--lyrics_txt", f"{lyrics_txt_path}",
        "--run_n_segments", str(num_segments),
        "--stage2_batch_size", "4",
        "--output_dir", f"{output_dir}",
        "--cuda_idx", "0",
        "--max_new_tokens", str(max_new_tokens)
    ]

    # Set up environment variables for CUDA with optimized settings
    env = os.environ.copy()
    env.update({
        "CUDA_VISIBLE_DEVICES": "0",
        "CUDA_HOME": "/usr/local/cuda",
        "PATH": f"/usr/local/cuda/bin:{env.get('PATH', '')}",
        "LD_LIBRARY_PATH": f"/usr/local/cuda/lib64:{env.get('LD_LIBRARY_PATH', '')}"
    })
    
    # Execute the command
    try:
        subprocess.run(command, check=True, env=env)
        print("Command executed successfully!")
        
        # Check and print the contents of the output folder
        output_files = os.listdir(output_dir)
        if output_files:
            print("Output folder contents:")
            for file in output_files:
                print(f"- {file}")

            last_mp3 = get_last_mp3_file(output_dir)

            if last_mp3:
                print("Last .mp3 file:", last_mp3)
                instrumental_mp3_path = "./output/vocoder/stems/instrumental.mp3"
                vocal_mp3_path = "./output/vocoder/stems/vocal.mp3"
                return last_mp3, instrumental_mp3_path, vocal_mp3_path
            else:
                return None, None, None
        else:
            print("Output folder is empty.")
            raise gr.Error(f"Error occurred: Output folder is empty.")
    except subprocess.CalledProcessError as e:
        print(f"Error occurred: {e}")
        raise gr.Error(f"Error occurred: {e}")
    finally:
        # Clean up temporary files
        os.remove(genre_txt_path)
        os.remove(lyrics_txt_path)
        print("Temporary files deleted.")

# Gradio 

with gr.Blocks() as demo:
    with gr.Column():
        gr.Markdown("# YuE: Open Music Foundation Models for Full-Song Generation")
        gr.HTML("""
        <div style="display:flex;column-gap:4px;">
            <a href="https://github.com/multimodal-art-projection/YuE">
                <img src='https://img.shields.io/badge/GitHub-Repo-blue'>
            </a> 
            <a href="https://map-yue.github.io">
                <img src='https://img.shields.io/badge/Project-Page-green'>
            </a>
            <a href="https://huggingface.co/spaces/fffiloni/YuE?duplicate=true">
                <img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/duplicate-this-space-sm.svg" alt="Duplicate this Space">
            </a>
        </div>
        """)
        with gr.Row():
            with gr.Column():
                with gr.Accordion("Pro Tips", open=False):
                    gr.Markdown(f"""
                        **Tips:**
                        1. `genres` should include details like instruments, genre, mood, vocal timbre, and vocal gender.
                        2. The length of `lyrics` segments and the `--max_new_tokens` value should be matched. For example, if `--max_new_tokens` is set to 3000, the maximum duration for a segment is around 30 seconds. Ensure your lyrics fit this time frame.
                        
                            
                        **Notice:**
                        1. A suitable [Genre] tag consists of five components: genre, instrument, mood, gender, and timbre. All five should be included if possible, separated by spaces. The values of timbre should include "vocal" (e.g., "bright vocal").

                        2. Although our tags have an open vocabulary, we have provided the 200 most commonly used <a href="https://github.com/multimodal-art-projection/YuE/blob/main/top_200_tags.json" id="tags_link" target="_blank">tags</a>. It is recommended to select tags from this list for more stable results.

                        3. The order of the tags is flexible. For example, a stable genre control string might look like: "inspiring female uplifting pop airy vocal electronic bright vocal vocal."

                        4. Additionally, we have introduced the "Mandarin" and "Cantonese" tags to distinguish between Mandarin and Cantonese, as their lyrics often share similarities.
                        """)
                genre_txt = gr.Textbox(
                    label="Genre", 
                    placeholder="Example: inspiring female uplifting pop airy vocal...",
                    info="Text containing genre tags that describe the musical style or characteristics (e.g., instrumental, genre, mood, vocal timbre, vocal gender). This is used as part of the generation prompt."
                )
                lyrics_txt = gr.Textbox(
                    label="Lyrics", lines=12,
                    placeholder="Type the lyrics here...",
                    info="Text containing the lyrics for the music generation. These lyrics will be processed and split into structured segments to guide the generation process."
                )
                
            with gr.Column():
                
                num_segments = gr.Number(label="Number of Segments", value=2, interactive=False)
                max_new_tokens = gr.Slider(label="Max New Tokens", minimum=500, maximum="3000", step=500, value=1500, interactive=True)
                
                submit_btn = gr.Button("Submit")
                music_out = gr.Audio(label="Audio Result")
                with gr.Accordion("Vocal & Instrumental", open=False):
                    instrumental = gr.Audio(label="Intrumental")
                    vocal = gr.Audio(label="Vocal")

        gr.Examples(
            examples = [
                [
                    "female blues airy vocal bright vocal piano sad romantic guitar jazz",
                    """[verse]
In the quiet of the evening, shadows start to fall
Whispers of the night wind echo through the hall
Lost within the silence, I hear your gentle voice
Guiding me back homeward, making my heart rejoice

[chorus]
Don't let this moment fade, hold me close tonight
With you here beside me, everything's alright
Can't imagine life alone, don't want to let you go
Stay with me forever, let our love just flow"""
                ],
                [
                    "rap piano street tough piercing vocal hip-hop synthesizer clear vocal male",
                    """[verse]
Woke up in the morning, sun is shining bright
Chasing all my dreams, gotta get my mind right
City lights are fading, but my vision's clear
Got my team beside me, no room for fear
Walking through the streets, beats inside my head
Every step I take, closer to the bread
People passing by, they don't understand
Building up my future with my own two hands

[chorus]
This is my life, and I'm aiming for the top
Never gonna quit, no, I'm never gonna stop
Through the highs and lows, I'mma keep it real
Living out my dreams with this mic and a deal"""
                ]
            ], 
             inputs = [genre_txt, lyrics_txt]
        )
    
    submit_btn.click(
        fn = infer, 
        inputs = [genre_txt, lyrics_txt, num_segments, max_new_tokens],
        outputs = [music_out, instrumental, vocal]
    )
demo.queue().launch(show_api=False, show_error=True)