import requests
import gradio as gr
import os
import torch
import json
import time
from transformers import AutoTokenizer, AutoModelForCausalLM

# Check if CUDA is available and set the device accordingly
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# API URLs and headers
AUDIO_API_URL = "https://api-inference.huggingface.co/models/MIT/ast-finetuned-audioset-10-10-0.4593"
JANUS_API_URL = "https://api-inference.huggingface.co/models/deepseek-ai/Janus-1.3B"
headers = {"Authorization": f"Bearer {os.environ.get('HF_TOKEN')}"}

def format_error(message):
    """Helper function to format error messages as JSON"""
    return {"error": message}

def create_lyrics_prompt(classification_results):
    """Create a prompt for lyrics generation based on classification results"""
    # Get the top genre and its characteristics
    top_result = classification_results[0]
    genre = top_result['label']
    confidence = float(top_result['score'].strip('%')) / 100

    # Create a detailed prompt
    prompt = f"""Write song lyrics in the style of {genre} music. The song should capture the essence of this genre.
    Additional musical elements detected: {', '.join(r['label'] for r in classification_results[1:3])}
    
    Please write creative and original lyrics that:
    1. Match the {genre} style
    2. Have a clear structure (verse, chorus)
    3. Reflect the mood and themes common in this genre
    
    Generate the lyrics:
    """
    return prompt

def generate_lyrics_with_retry(prompt, max_retries=5, initial_wait=2):
    """Generate lyrics using the Janus model with retry logic"""
    wait_time = initial_wait
    
    for attempt in range(max_retries):
        try:
            response = requests.post(
                JANUS_API_URL,
                headers=headers,
                json={
                    "inputs": prompt,
                    "parameters": {
                        "max_new_tokens": 200,
                        "temperature": 0.7,
                        "top_p": 0.9,
                        "return_full_text": False
                    }
                }
            )
            
            if response.status_code == 200:
                return response.json()[0]["generated_text"]
            elif response.status_code == 503:
                print(f"Model loading, attempt {attempt + 1}/{max_retries}. Waiting {wait_time} seconds...")
                time.sleep(wait_time)
                wait_time *= 1.5  # Increase wait time for next attempt
                continue
            else:
                return f"Error generating lyrics: {response.text}"
                
        except Exception as e:
            if attempt == max_retries - 1:  # Last attempt
                return f"Error after {max_retries} attempts: {str(e)}"
            time.sleep(wait_time)
            wait_time *= 1.5
    
    return "Failed to generate lyrics after multiple attempts. Please try again."

def format_results(classification_results, lyrics, prompt):
    """Format the results for display"""
    # Format classification results
    classification_text = "Classification Results:\n"
    for i, result in enumerate(classification_results):
        classification_text += f"{i+1}. {result['label']}: {result['score']}\n"
    
    # Format final output
    output = f"""
{classification_text}
\n---Generated Lyrics---\n
{lyrics}
"""
    return output

def classify_and_generate(audio_file):
    """
    Classify the audio and generate matching lyrics
    """
    if audio_file is None:
        return "Please upload an audio file."
    
    try:
        token = os.environ.get('HF_TOKEN')
        if not token:
            return "Error: HF_TOKEN environment variable is not set. Please set your Hugging Face API token."
        
        # First, classify the audio
        with open(audio_file, "rb") as f:
            data = f.read()
        
        print("Sending request to Audio Classification API...")
        response = requests.post(AUDIO_API_URL, headers=headers, data=data)
        
        if response.status_code == 200:
            classification_results = response.json()
            # Format classification results
            formatted_results = []
            for result in classification_results:
                formatted_results.append({
                    'label': result['label'],
                    'score': f"{result['score']*100:.2f}%"
                })
            
            # Generate lyrics based on classification with retry logic
            print("Generating lyrics based on classification...")
            prompt = create_lyrics_prompt(formatted_results)
            lyrics = generate_lyrics_with_retry(prompt)
            
            # Format and return results
            return format_results(formatted_results, lyrics, prompt)
            
        elif response.status_code == 401:
            return "Error: Invalid or missing API token. Please check your Hugging Face API token."
        elif response.status_code == 503:
            return "Error: Model is loading. Please try again in a few seconds."
        else:
            return f"Error: API returned status code {response.status_code}\nResponse: {response.text}"
            
    except Exception as e:
        import traceback
        error_details = traceback.format_exc()
        return f"Error processing request: {str(e)}\nDetails:\n{error_details}"

# Create Gradio interface
iface = gr.Interface(
    fn=classify_and_generate,
    inputs=gr.Audio(type="filepath", label="Upload Audio File"),
    outputs=gr.Textbox(
        label="Results",
        lines=15,
        placeholder="Upload an audio file to see classification results and generated lyrics..."
    ),
    title="Music Genre Classifier + Lyric Generator",
    description="Upload an audio file to classify its genre and generate matching lyrics using AI.",
    examples=[],
)

# Launch the interface
if __name__ == "__main__":
    iface.launch(server_name="0.0.0.0", server_port=7860)