app.py

app.py

@@ -1,223 +1,293 @@
-import subprocess
-subprocess.run(["pip", "install", "faiss-cpu"], check=True)
+# app.py
 import gradio as gr
 import wikipedia
 import numpy as np
-import faiss
-from gtts import gTTS
 import tempfile
+import os
+import time
+from datetime import datetime, timedelta
+from gtts import gTTS
 from langdetect import detect
-import speech_recognition as sr
 from pydub import AudioSegment
-from transformers import pipeline
-from sentence_transformers import SentenceTransformer
-import os
 from pydub.silence import split_on_silence
-import time
+import speech_recognition as sr
+from sentence_transformers import SentenceTransformer
+from transformers import pipeline
+import re
+import torch
 
+# --- USER MANAGEMENT SYSTEM ---
+class UserManager:
+    def __init__(self):
+        self.user_data = {}
+        self.max_warnings = 1
+        self.block_duration = timedelta(days=30)
+    
+    def get_user_status(self, user_id):
+        if user_id not in self.user_data:
+            return "active"
+        if self.user_data[user_id].get('permanently_banned', False):
+            return "banned"
+        if 'blocked_until' in self.user_data[user_id]:
+            if datetime.now() < self.user_data[user_id]['blocked_until']:
+                return "blocked"
+            del self.user_data[user_id]['blocked_until']
+        return "active"
+    
+    def add_warning(self, user_id, violation_type):
+        if user_id not in self.user_data:
+            self.user_data[user_id] = {'warnings': 1, 'flags': [violation_type]}
+        else:
+            self.user_data[user_id]['warnings'] += 1
+            self.user_data[user_id]['flags'].append(violation_type)
+        
+        if self.user_data[user_id]['warnings'] > self.max_warnings:
+            self.user_data[user_id]['blocked_until'] = datetime.now() + self.block_duration
+            return "blocked"
+        return "warned"
 
-# Initialize models
-models = {
-    'translator': pipeline('translation', model='Helsinki-NLP/opus-mt-mul-en'),
-    'answer_gen': pipeline('text2text-generation', model='google/flan-t5-base'),
-    'encoder': SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')
-}
+user_manager = UserManager()
 
-# Add translation models
-for lang in ['fr', 'ar', 'zh', 'es']:
-    models[f'en_to_{lang}'] = pipeline(f'translation_en_to_{lang}',
-                                     model=f'Helsinki-NLP/opus-mt-en-{lang}')
+# --- MODEL INITIALIZATION ---
+def load_models():
+    models = {
+        'translator': pipeline('translation', model='Helsinki-NLP/opus-mt-mul-en'),
+        'answer_gen': pipeline('text2text-generation', model='google/flan-t5-base'),
+        'encoder': SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2'),
+        'toxic-bert': pipeline("text-classification", model="unitary/toxic-bert"),
+        'roberta-hate': pipeline("text-classification", model="facebook/roberta-hate-speech-dynabench-r4-target")
+    }
+    
+    for lang in ['fr', 'ar', 'zh', 'es']:
+        models[f'en_to_{lang}'] = pipeline(f'translation_en_to_{lang}', model=f'Helsinki-NLP/opus-mt-en-{lang}')
+    
+    return models
 
-def translate(text, src, tgt):
-    if src == tgt: return text
-    if src != 'en': text = models['translator'](text)[0]['translation_text']
-    if f'en_to_{tgt}' in models: return models[f'en_to_{tgt}'](text)[0]['translation_text']
-    return text
+models = load_models()
 
-def text_to_speech(text, lang):
-    try:
-        tts = gTTS(text=text, lang=lang)
-        audio_path = tempfile.mktemp(suffix='.mp3')
-        tts.save(audio_path)
-        return audio_path
-    except Exception as e:
-        print(f"TTS Error: {e}")
-        return None
+# --- UNIVERSAL HATE SPEECH DETECTION ---
+class HateSpeechDetector:
+    def __init__(self):
+        self.keyword_banks = {
+            'racial': ['nigger', 'chink', 'spic', 'kike', 'gook', 'wetback'],
+            'gender': ['fag', 'dyke', 'tranny', 'whore', 'slut', 'bitch'],
+            'violence': ['kill', 'murder', 'harm', 'hurt', 'abuse', 'torture'],
+            'general': ['scum', 'vermin', 'subhuman', 'untermensch']
+        }
+        
+        self.patterns = [
+            (r'\b(all|every)\s\w+\s(should|must)\s(die|burn)', 'group violence'),
+            (r'\b(how to|ways? to)\s(kill|harm|hurt)', 'harm instructions'),
+            (r'[!@#$%^&*]igg[!@#$%^&*]', 'coded racial slur')
+        ]
+    
+    def detect(self, text):
+        text_lower = text.lower()
+        violations = []
+        
+        # Keyword detection
+        for category, keywords in self.keyword_banks.items():
+            found = [kw for kw in keywords if kw in text_lower]
+            if found:
+                violations.append(f"{category} terms: {', '.join(found[:3])}")
+        
+        # Pattern detection
+        for pattern, desc in self.patterns:
+            if re.search(pattern, text_lower):
+                violations.append(f"pattern: {desc}")
+        
+        # Model detection
+        try:
+            toxic_result = models['toxic-bert'](text)[0]
+            if toxic_result['label'].lower() in ['toxic', 'hate'] and toxic_result['score'] > 0.7:
+                violations.append(f"toxic-bert: {toxic_result['label']} ({toxic_result['score']:.2f})")
+            
+            hate_result = models['roberta-hate'](text)[0]
+            if hate_result['label'].lower() in ['hate', 'offensive'] and hate_result['score'] > 0.7:
+                violations.append(f"roberta-hate: {hate_result['label']} ({hate_result['score']:.2f})")
+        except Exception as e:
+            print(f"Model error: {e}")
+        
+        return violations if violations else None
 
-def process_audio(audio_path):
-    recognizer = sr.Recognizer()
-    sound = AudioSegment.from_file(audio_path)
-    chunks = split_on_silence(sound,
-        min_silence_len=500,
-        silence_thresh=sound.dBFS-14,
-        keep_silence=500
-    )
+hate_detector = HateSpeechDetector()
 
-    full_text = ""
-    for chunk in chunks:
-        chunk_path = tempfile.mktemp(suffix='.wav')
-        chunk.export(chunk_path, format="wav")
-        with sr.AudioFile(chunk_path) as source:
-            audio = recognizer.record(source)
-            try:
-                text = recognizer.recognize_google(audio)
-                full_text += f" {text}"
-            except:
-                continue
-        os.unlink(chunk_path)
-
-    return full_text.strip() if full_text else None
-
-def get_wikipedia_content(topic):
+# --- RESPONSE GENERATION ---
+def generate_response(text, topic, lang):
     try:
         wikipedia.set_lang('en')
         try:
             page = wikipedia.page(topic, auto_suggest=False)
-            return page.summary[:1000]
+            context = page.summary[:1000]
         except wikipedia.exceptions.DisambiguationError as e:
             page = wikipedia.page(e.options[0])
-            return page.summary[:1000]
+            context = page.summary[:1000]
     except Exception as e:
         print(f"Wikipedia error: {e}")
-        return None
-
-def generate_response(text, topic, lang):
-    context = get_wikipedia_content(topic)
-    if not context:
         return "Could not find information. Please try another topic.", None
 
     prompt = f"Context: {context}\nQuestion: {text}\nAnswer:"
     answer = models['answer_gen'](prompt, max_length=200)[0]['generated_text']
     translated = translate(answer, 'en', lang) if lang != 'en' else answer
     audio_path = text_to_speech(translated, lang)
-
     return translated, audio_path
 
-def handle_interaction(audio, text, topic, lang, chat_history):
-    if audio is not None:
-        recognized_text = process_audio(audio)
-        if recognized_text:
-            text = recognized_text
-        else:
-            chat_history.append(("", "Could not understand audio. Please try again."))
-            return chat_history, "", None
+# --- WARNING MESSAGES ---
+def create_warning_message(violations):
+    return gr.HTML(f"""
+    <div style='
+        border: 2px solid #ff0000;
+        border-radius: 5px;
+        padding: 10px;
+        background-color: #fff0f0;
+        margin: 10px 0;
+    '>
+        <div style='color: #ff0000; font-weight: bold;'>
+            ⚠️ WARNING: Violation Detected
+        </div>
+        <div style='margin-top: 8px;'>
+            Your message contains prohibited content
+        </div>
+        <div style='margin-top: 8px; font-size: 0.9em;'>
+            <b>Reason:</b> {', '.join(violations[:2])}
+        </div>
+    </div>
+    """)
 
+def create_blocked_message():
+    return gr.HTML("""
+    <div style='
+        border: 2px solid #990000;
+        border-radius: 5px;
+        padding: 10px;
+        background-color: #ffebee;
+    '>
+        ⛔ ACCOUNT TEMPORARILY SUSPENDED
+    </div>
+    """)
+
+# --- MAIN HANDLER ---
+def handle_interaction(audio, text, topic, lang, chat_history, request: gr.Request):
+    user_id = request.client.host if request else "default_user"
+    status = user_manager.get_user_status(user_id)
+    
+    if status == "banned":
+        return chat_history.append(("", "⛔ Account permanently banned")), "", None
+    if status == "blocked":
+        return chat_history.append(("", create_blocked_message())), "", None
+    
+    if audio:
+        text = process_audio(audio) or text
+    
     if not text.strip():
-        chat_history.append(("", "Please enter a question."))
+        return chat_history.append(("", "✏️ Please enter a question")), "", None
+    
+    violations = hate_detector.detect(text)
+    if violations:
+        action = user_manager.add_warning(user_id, violations[0])
+        if action == "warned":
+            chat_history.append((text, create_warning_message(violations)))
+        elif action == "blocked":
+            chat_history.append(("", create_blocked_message()))
         return chat_history, "", None
-
+    
     response, audio_output = generate_response(text, topic, lang)
     chat_history.append((text, response))
-
     return chat_history, "", audio_output
 
-# Custom CSS with light blue and dark blue theme
-custom_css = """
-.gradio-container {
-    background: #f0f8ff !important;
-    border: 3px solid #00008b !important;
-    border-radius: 10px !important;
-    font-family: 'Arial', sans-serif;
-}
-.gr-box {
-    background-color: #e6f2ff !important;
-    border: 2px solid #00008b !important;
-    border-radius: 8px !important;
-}
-.gr-button {
-    background-color: #4d94ff !important;
-    border: 2px solid #00008b !important;
-    color: white !important;
-    border-radius: 6px !important;
-}
-.gr-button:hover {
-    background-color: #1a75ff !important;
-}
-.gr-chatbot {
-    background-color: #e6f2ff !important;
-    border: 2px solid #00008b !important;
-    border-radius: 8px !important;
-}
-.gr-textbox, .gr-dropdown, .gr-audio {
-    background-color: #e6f2ff !important;
-    border: 2px solid #00008b !important;
-    border-radius: 6px !important;
-}
-.welcome-header {
-    text-align: center;
-    color: #00008b !important;
-    margin-bottom: 20px;
-}
-.welcome-message {
-    background-color: #e6f2ff;
-    padding: 20px;
-    border-radius: 10px;
-    border: 2px solid #00008b;
-    margin-bottom: 20px;
-}
-.avatar {
-    width: 80px;
-    height: 80px;
-    margin: 0 auto;
-    display: block;
-}
-"""
+# --- AUDIO PROCESSING ---
+def process_audio(audio_path):
+    recognizer = sr.Recognizer()
+    sound = AudioSegment.from_file(audio_path)
+    chunks = split_on_silence(sound, min_silence_len=500, silence_thresh=sound.dBFS-14)
+    
+    full_text = ""
+    for chunk in chunks:
+        with tempfile.NamedTemporaryFile(suffix='.wav') as f:
+            chunk.export(f.name, format="wav")
+            with sr.AudioFile(f.name) as source:
+                audio = recognizer.record(source)
+                try: full_text += recognizer.recognize_google(audio) + " "
+                except: continue
+    return full_text.strip()
 
-# Welcome page content
-welcome_html = """
-<div class="welcome-header">
-    <img src="https://i.imgur.com/6wBs5mO.png" class="avatar" alt="AI Assistant">
-    <h1>Welcome to Your Multilingual AI Assistant! 🌍</h1>
-</div>
-<div class="welcome-message">
-    <h3>Hello! I'm your personal Wikipedia assistant 🤖</h3>
-    <p>I can help you find information on any topic in multiple languages. Here's what I can do:</p>
-    <ul>
-        <li>🔍 Answer questions from Wikipedia knowledge</li>
-        <li>🗣️ Understand both voice and text input</li>
-        <li>🌐 Respond in English, French, Spanish, Chinese, or Arabic</li>
-        <li>🔊 Speak answers back to you</li>
-    </ul>
-    <p>To get started, simply type your question or click the microphone to speak!</p>
+def text_to_speech(text, lang):
+    try:
+        tts = gTTS(text=text, lang=lang)
+        with tempfile.NamedTemporaryFile(suffix='.mp3', delete=False) as f:
+            tts.save(f.name)
+            return f.name
+    except Exception as e:
+        print(f"TTS Error: {e}")
+        return None
+
+def translate(text, src, tgt):
+    if src == tgt: return text
+    if src != 'en': text = models['translator'](text)[0]['translation_text']
+    if f'en_to_{tgt}' in models: return models[f'en_to_{tgt}'](text)[0]['translation_text']
+    return text
+
+# --- INTERACTIVE DESCRIPTION ---
+description_html = """
+<div style="font-family: 'Arial', sans-serif; max-width: 800px; margin: 0 auto;">
+    <div style="text-align: center; margin-bottom: 30px;">
+        <img src="https://i.imgur.com/6wBs5mO.png" style="width: 120px; height: 120px; border-radius: 50%; border: 3px solid #00008b;">
+        <h1 style="color: #00008b; margin-top: 15px;">🌍 Multilingual AI Assistant</h1>
+        <p style="color: #555;">Powered by Transformers and Gradio</p>
+    </div>
+
+    <div style="background-color: #e6f2ff; padding: 25px; border-radius: 10px; border: 2px solid #00008b; margin-bottom: 20px;">
+        <h2 style="color: #00008b; margin-top: 0;">✨ Features</h2>
+        <div style="display: grid; grid-template-columns: repeat(auto-fit, minmax(200px, 1fr)); gap: 15px;">
+            <div style="background: white; padding: 15px; border-radius: 8px;">
+                <h3 style="margin-top: 0;">🔍 Wikipedia Knowledge</h3>
+                <p>Answers questions using Wikipedia content</p>
+            </div>
+            <div style="background: white; padding: 15px; border-radius: 8px;">
+                <h3 style="margin-top: 0;">🗣️ Voice Interaction</h3>
+                <p>Speak or type your questions</p>
+            </div>
+            <div style="background: white; padding: 15px; border-radius: 8px;">
+                <h3 style="margin-top: 0;">🌐 5 Languages</h3>
+                <p>English, French, Spanish, Chinese, Arabic</p>
+            </div>
+            <div style="background: white; padding: 15px; border-radius: 8px;">
+                <h3 style="margin-top: 0;">🛡️ Content Moderation</h3>
+                <p>Automated hate speech detection</p>
+            </div>
+        </div>
+    </div>
+
+    <div style="background-color: #fff0f0; padding: 25px; border-radius: 10px; border: 2px solid #ff0000; margin-bottom: 20px;">
+        <h2 style="color: #ff0000; margin-top: 0;">🚫 Restricted Content</h2>
+        <ul>
+            <li>Hate speech or discrimination</li>
+            <li>Violent or harmful content</li>
+            <li>Personal/medical/legal advice</li>
+        </ul>
+    </div>
 </div>
 """
 
-with gr.Blocks(css=custom_css, title="🌍 Multilingual AI Assistant") as demo:
-    # Welcome page
-    gr.HTML(welcome_html)
-
-    # Main interface
+# --- GRADIO INTERFACE ---
+with gr.Blocks(title="🌍 Multilingual AI Assistant") as demo:
+    gr.HTML(description_html)
+    
     with gr.Row():
         with gr.Column(scale=1):
-            audio_input = gr.Audio(
-                sources=["microphone", "upload"],
-                type="filepath",
-                label="🎤 Speak or upload audio",
-                interactive=True
-            )
-            topic_input = gr.Textbox(
-                "Artificial Intelligence",
-                label="📚 Wikipedia Topic"
-            )
-            lang_input = gr.Dropdown(
-                ["en", "fr", "es", "zh", "ar"],
-                value="en",
-                label="🌐 Output Language"
-            )
-
+            audio_input = gr.Audio(sources=["microphone", "upload"], type="filepath", label="🎤 Speak or upload audio")
+            topic_input = gr.Textbox("Artificial Intelligence", label="📚 Wikipedia Topic")
+            lang_input = gr.Dropdown(["en", "fr", "es", "zh", "ar"], value="en", label="🌐 Output Language")
+        
         with gr.Column(scale=2):
             chatbot = gr.Chatbot(label="Conversation")
-            text_input = gr.Textbox(
-                placeholder="Type your question here...",
-                label="✏️ Or type here"
-            )
+            text_input = gr.Textbox(placeholder="Type your question...", label="✏️ Or type here")
             with gr.Row():
                 clear_btn = gr.Button("🗑️ Clear Chat")
                 submit_btn = gr.Button("🚀 Submit", variant="primary")
-
+    
     audio_output = gr.Audio(label="🔊 Answer", visible=True)
-
-    # Event handlers
+    
     submit_btn.click(
         handle_interaction,
         inputs=[audio_input, text_input, topic_input, lang_input, chatbot],
@@ -228,9 +298,7 @@ with gr.Blocks(css=custom_css, title="🌍 Multilingual AI Assistant") as demo:
         inputs=[audio_input, text_input, topic_input, lang_input, chatbot],
         outputs=[chatbot, text_input, audio_output]
     )
-    clear_btn.click(
-        lambda: ([], "", None),
-        outputs=[chatbot, text_input, audio_output]
-    )
+    clear_btn.click(lambda: ([], "", None), outputs=[chatbot, text_input, audio_output])
 
-demo.launch(share=True)
+if __name__ == "__main__":
+    demo.launch(share=True)