script_analyzer.py

import gradio as gr
from transformers import pipeline, AutoModelForSequenceClassification, AutoTokenizer, AutoModelForCausalLM
import matplotlib.pyplot as plt

# Load sentiment and generation models as previously set up
tokenizer_sentiment = AutoTokenizer.from_pretrained("cardiffnlp/twitter-roberta-base-sentiment")
model_sentiment = AutoModelForSequenceClassification.from_pretrained("cardiffnlp/twitter-roberta-base-sentiment")
sentiment_classifier = pipeline("sentiment-analysis", model=model_sentiment, tokenizer=tokenizer_sentiment)

tokenizer_gpt = AutoTokenizer.from_pretrained("EleutherAI/gpt-neo-2.7B")
model_gpt = AutoModelForCausalLM.from_pretrained("EleutherAI/gpt-neo-2.7B")

# Updated generate_text function with padding token fix
def generate_text(prompt):
    # Add '[PAD]' as a padding token if not set
    if tokenizer_gpt.pad_token is None:
        tokenizer_gpt.add_special_tokens({'pad_token': '[PAD]'})

    # Prepare the input tensors with attention_mask and padding
    inputs = tokenizer_gpt(prompt, return_tensors="pt", padding=True, truncation=True, max_length=50)
    # Generate text using max_new_tokens instead of max_length
    outputs = model_gpt.generate(
        inputs["input_ids"],
        attention_mask=inputs["attention_mask"],
        max_new_tokens=50,
        num_return_sequences=1,
        no_repeat_ngram_size=2,
        pad_token_id=tokenizer_gpt.pad_token_id  # Ensure pad_token_id is set
    )
    generated_text = tokenizer_gpt.decode(outputs[0], skip_special_tokens=True)
    return generated_text.strip()

# Function to analyze each line of the script
def analyze_script(script):
    global all_scores, descriptions, music_cues
    lines = script.strip().split("\n")
    all_scores = []
    descriptions = []
    music_cues = []
    analysis_results = []
    
    for i, line in enumerate(lines):
        result = sentiment_classifier(line)[0]
        sentiment = result['label']
        score = result['score']
        
        description_prompt = f"Describe a scene with the sentiment '{sentiment}' for the line: '{line}'"
        description = generate_text(description_prompt)
        
        music_cue_prompt = f"Suggest music elements (like tempo, key, and instrumentation) that would fit a scene with the sentiment '{sentiment}': '{line}'"
        music_cue = generate_text(music_cue_prompt)
        
        all_scores.append(score)
        descriptions.append(description)
        music_cues.append(music_cue)
        
        analysis_results.append(
            {
                "Line": f"Line {i + 1}: {line}",
                "Sentiment": f"{sentiment} (Score: {round(score, 2)})",
                "Description Suggestion": description,
                "Music Cue": music_cue
            }
        )
    
    graph_path = generate_script_graph()
    return analysis_results, graph_path

# Generate the emotional arc graph with music cues for the entire script
def generate_script_graph():
    plt.figure(figsize=(12, 6))
    plt.plot(all_scores, marker='o', linestyle='-', color='b', label='Sentiment Intensity')
    
    for i, score in enumerate(all_scores):
        plt.text(i, score, music_cues[i], fontsize=8, ha='right', rotation=45)
    
    plt.title('Emotional and Musical Arc for Entire Script')
    plt.xlabel('Script Lines (Accumulative)')
    plt.ylabel('Sentiment Intensity')
    plt.legend()
    plt.tight_layout()
    
    plot_path = "script_emotional_arc.png"
    plt.savefig(plot_path)
    plt.close()
    return plot_path

# Custom Gradio component to display dashboard results with icons
def format_dashboard(results):
    formatted_results = ""
    for result in results:
        formatted_results += f"""
        <div class="dashboard-box">
            <p><img src="https://geist-ui.dev/icons/activity.svg" class="dashboard-icon" alt="Line icon"> <strong>{result['Line']}</strong></p>
            <p><img src="https://geist-ui.dev/icons/bar-chart.svg" class="dashboard-icon" alt="Sentiment icon"> <strong>Sentiment:</strong> {result['Sentiment']}</p>
            <p><img src="https://geist-ui.dev/icons/eye.svg" class="dashboard-icon" alt="Description icon"> <strong>Description Suggestion:</strong> {result['Description Suggestion']}</p>
            <p><img src="https://geist-ui.dev/icons/music.svg" class="dashboard-icon" alt="Music Cue icon"> <strong>Music Cue:</strong> {result['Music Cue']}</p>
        </div>
        """
    return formatted_results

# Gradio interface to analyze script and display the dashboard
with gr.Blocks(css="custom.css") as interface:
    gr.Markdown("## Script Sentiment and Music Cue Analyzer", elem_id="title")
    gr.Markdown("Enter your script line-by-line, and this tool will analyze sentiment, generate scene descriptions, suggest music cues, and show an emotional and musical arc.", elem_id="description")
    
    script_input = gr.Textbox(lines=10, placeholder="Enter your script here, one line per thought or dialogue.", label="Script")
    display_dashboard_button = gr.Button("Analyze Script")
    
    output_dashboard = gr.HTML(label="Dashboard Results")
    output_graph = gr.Image(label="Emotional and Musical Arc for Entire Script")

    def display_dashboard(script):
        analysis_results, graph_path = analyze_script(script)
        dashboard_content = format_dashboard(analysis_results)
        return dashboard_content, graph_path
    
    display_dashboard_button.click(display_dashboard, inputs=script_input, outputs=[output_dashboard, output_graph])

interface.launch()