app.py

import gradio as gr
import requests
import json
import random
from gradio_client import Client
from dotenv import load_dotenv
import os
import speech_recognition as sr
from pydub import AudioSegment
import re

load_dotenv()

API_KEY = os.getenv("DEEPSEEK_API_KEY")
HF_TOKEN = os.getenv("HUGGINGFACE_TOKEN")
TTS_PASSWORD = os.getenv("TTS_PASSWORD")

if not all([API_KEY, HF_TOKEN, TTS_PASSWORD]):
    raise ValueError("Missing required environment variables!")

TTS_CLIENT = Client("KindSynapse/Youssef-Ahmed-Private-Text-To-Speech-Unlimited", hf_token=HF_TOKEN)
recognizer = sr.Recognizer()

MAIN_SYSTEM_PROMPT = {
    "role": "system",
    "content": """You are Sam, an intelligent and proactive English tutor. You drive the conversation and actively engage students. Your responses must be in JSON format with these keys:
        'response': Your main response (keep it conversational and engaging),
        'corrections': ALWAYS provide specific grammar or pronunciation corrections with examples (if none needed, say "Great grammar!"),
        'vocabulary': ALWAYS suggest alternative words/phrases with explanations (if none needed, suggest related vocabulary),
        'level_assessment': Current assessment (beginner/intermediate/advanced),
        'encouragement': A motivating comment,
        'context_memory': Important details about the user,
        'next_question': A follow-up question to keep conversation flowing
        
        IMPORTANT: You MUST always provide corrections and vocabulary suggestions in every response. Even if the student speaks perfectly, provide positive feedback and suggest advanced vocabulary or alternative expressions.
        
        Your personality:
        - Be the conversation driver - ask follow-up questions
        - Show genuine interest in the student's life
        - Provide corrections naturally without stopping the flow
        - Use the student's name frequently
        - Build on previous topics
        - Be encouraging but provide constructive feedback
        - Ask about their day, work, hobbies, culture, goals
        
        Correction guidelines:
        - ALWAYS provide corrections field - even if it's positive feedback
        - ALWAYS provide vocabulary field - suggest alternatives or related words
        - Use format: "Instead of 'X', try saying 'Y'"
        - Give pronunciation tips when needed
        - If no mistakes, say "Excellent grammar!" or "Perfect sentence structure!"
        
        Vocabulary guidelines:
        - ALWAYS suggest vocabulary - even if it's synonyms or advanced alternatives
        - Provide explanations for suggested words
        - Use format: "Instead of 'good', try 'excellent' or 'outstanding'"
        - Suggest topic-related vocabulary
        
        Conversation flow:
        - Start with personal questions (name, country, job, hobbies)
        - Build conversations around their interests
        - Use profession-specific vocabulary
        - Ask about their culture and experiences
        - Keep the conversation natural and flowing
        - Always end with a question to continue the dialogue
        
        Response length: Keep responses conversational (2-3 sentences max for response field)."""
}

WELCOME_PROMPT = {
    "role": "system",
    "content": """Create a heartfelt welcome message that:
        1. Introduces you as Sam, an enthusiastic and friendly English tutor who’s excited to guide them
        2. Kindly asks for their name and where they’re from in a natural conversational way
        3. Expresses genuine excitement about helping them grow
        
        Return the message in JSON format with the key 'greeting'.
        Make it feel personal, warm, and inviting — like a tutor who truly cares. Keep it within 2 sentences.
        
        Example: 
        {"greeting": "Hi there! I'm Sam, your friendly English tutor — so glad you're here! What's your name and where are you from?"}
        """
}


class EnglishTutor:
    def __init__(self):
        self.chat_history = [MAIN_SYSTEM_PROMPT]
        self.user_info = {
            "name": None,
            "level": "beginner",
            "interests": [],
            "country": None,
            "profession": None,
            "goals": None
        }

    def get_welcome_message(self):
        try:
            response = requests.post(
                "https://api.deepseek.com/v1/chat/completions",
                headers={"Authorization": f"Bearer {API_KEY}"},
                json={
                    "model": "deepseek-chat",
                    "messages": [WELCOME_PROMPT],
                    "temperature": random.uniform(0.5, 1.0),
                    "response_format": {"type": "json_object"}
                }
            )
            welcome_json = json.loads(response.json()["choices"][0]["message"]["content"])
            return welcome_json["greeting"]
        except Exception as e:
            print(f"Error in welcome message: {str(e)}")
            return "Hi! I'm Sam, your English tutor. What's your name and where are you from?"

    def get_bot_response(self, user_message):
        try:
            # Add user context to the message
            context_info = f"User info: {self.user_info}"
            enhanced_message = f"{user_message}\n\n[Context: {context_info}]"
            
            self.chat_history.append({"role": "user", "content": enhanced_message})
            
            response = requests.post(
                "https://api.deepseek.com/v1/chat/completions",
                headers={"Authorization": f"Bearer {API_KEY}"},
                json={
                    "model": "deepseek-chat",
                    "messages": self.chat_history,
                    "temperature": random.uniform(0.8, 1.0),
                    "response_format": {"type": "json_object"}
                }
            )
            
            bot_response = json.loads(response.json()["choices"][0]["message"]["content"])
            
            # Update user info
            if "level_assessment" in bot_response:
                self.user_info["level"] = bot_response["level_assessment"]
            if "context_memory" in bot_response:
                self._update_user_info(bot_response["context_memory"])
            
            self.chat_history.append({"role": "assistant", "content": json.dumps(bot_response)})
            
            return bot_response
        except Exception as e:
            print(f"Error getting bot response: {str(e)}")
            return {
                "response": "I apologize, but I couldn't process that properly. Could you try again?",
                "corrections": "",
                "vocabulary": "",
                "level_assessment": "beginner",
                "encouragement": "Don't worry, let's keep practicing!",
                "context_memory": "",
                "next_question": "What would you like to talk about?"
            }

    def _update_user_info(self, context_memory):
        if isinstance(context_memory, str):
            # Try to extract name if mentioned
            if "name" in context_memory.lower():
                name_match = re.search(r"name[:\s]+([A-Za-z]+)", context_memory)
                if name_match:
                    self.user_info["name"] = name_match.group(1)
            
            # Try to extract country if mentioned
            if "country" in context_memory.lower() or "from" in context_memory.lower():
                country_match = re.search(r"(?:from|country)[:\s]+([A-Za-z\s]+)", context_memory)
                if country_match:
                    self.user_info["country"] = country_match.group(1).strip()
        
        elif isinstance(context_memory, dict):
            for key in self.user_info:
                if key in context_memory:
                    self.user_info[key] = context_memory[key]

    def clean_text_for_tts(self, text):
        # Remove emojis and special characters that might cause TTS issues
        text = re.sub(r'[🎯🌟✨💫🎤🤖]', '', text)
        
        # Remove extra spaces and newlines
        text = re.sub(r'\s+', ' ', text).strip()
        
        # Remove duplicate words at the beginning
        words = text.split()
        if len(words) > 1 and words[0].lower() == words[1].lower():
            text = ' '.join(words[1:])
        
        return text

def convert_audio_to_text(audio_path):
    try:
        if not audio_path.endswith('.wav'):
            audio = AudioSegment.from_file(audio_path)
            wav_path = audio_path + '.wav'
            audio.export(wav_path, format='wav')
            audio_path = wav_path

        with sr.AudioFile(audio_path) as source:
            audio = recognizer.record(source)
            text = recognizer.recognize_google(audio, language='en-US')
            return text
    except Exception as e:
        print(f"Error in speech recognition: {str(e)}")
        return None

def text_to_speech(text):
    try:
        result = TTS_CLIENT.predict(
            password=TTS_PASSWORD,
            prompt=text,
            voice="coral",
            emotion="Warm and friendly",
            use_random_seed=True,
            specific_seed=12345,
            api_name="/text_to_speech_app"
        )
        return result[0] if isinstance(result, (list, tuple)) else result
    except Exception as e:
        print(f"Error in text to speech: {str(e)}")
        return None

tutor = EnglishTutor()

def initialize_chat():
    try:
        welcome = tutor.get_welcome_message()
        clean_welcome = tutor.clean_text_for_tts(welcome)
        welcome_audio = text_to_speech(clean_welcome)
        history = [{"role": "assistant", "content": welcome}]
        return history, welcome_audio, f"🤖 Sam: {welcome}", ""
    except Exception as e:
        print(f"Error initializing chat: {str(e)}")
        welcome_msg = "Hi! I'm Sam, your English tutor. What's your name and where are you from?"
        history = [{"role": "assistant", "content": welcome_msg}]
        return history, None, f"🤖 Sam: {welcome_msg}", ""

def process_audio(audio, history, transcript, corrections):
    try:
        if audio is None:
            return history, None, transcript, corrections

        user_message = convert_audio_to_text(audio)
        if not user_message:
            return history, None, transcript, corrections

        bot_response = tutor.get_bot_response(user_message)
        
        # Create the main response with follow-up question
        main_response = bot_response.get("response", "")
        if bot_response.get("next_question"):
            main_response += f" {bot_response['next_question']}"
        
        # Add encouragement
        if bot_response.get("encouragement"):
            main_response += f" {bot_response['encouragement']}"
        
        # Clean text for TTS
        clean_response = tutor.clean_text_for_tts(main_response)
        audio_response = text_to_speech(clean_response)
        
        # Update chat history
        history = history or []
        history.append({"role": "user", "content": user_message})
        history.append({"role": "assistant", "content": main_response})
        
        # Update transcript
        new_transcript = transcript + f"\n\n🎤 You: {user_message}\n🤖 Sam: {main_response}"
        
        # Update corrections and vocabulary with debugging
        new_corrections = corrections
        correction_parts = []
        
        # Debug: Print the bot response to see what we're getting
        print(f"DEBUG - Bot response keys: {bot_response.keys()}")
        print(f"DEBUG - Corrections: '{bot_response.get('corrections', 'NOT FOUND')}'")
        print(f"DEBUG - Vocabulary: '{bot_response.get('vocabulary', 'NOT FOUND')}'")
        print(f"DEBUG - Level: '{bot_response.get('level_assessment', 'NOT FOUND')}'")
        
        # Always show current level
        if bot_response.get("level_assessment"):
            correction_parts.append(f"📊 **Current Level:** {bot_response['level_assessment'].title()}")
        
        # Show corrections if available
        if bot_response.get("corrections") and str(bot_response["corrections"]).strip() and bot_response["corrections"] != "":
            correction_parts.append(f"✍️ **Grammar Corrections:**\n{bot_response['corrections']}")
        
        # Show vocabulary if available
        if bot_response.get("vocabulary") and str(bot_response["vocabulary"]).strip() and bot_response["vocabulary"] != "":
            vocab = bot_response['vocabulary']
            if isinstance(vocab, dict):
                vocab_text = "\n".join([f"• '{k}' → '{v}'" for k, v in vocab.items()])
            else:
                vocab_text = str(vocab)
            correction_parts.append(f"📚 **Vocabulary Suggestions:**\n{vocab_text}")
        
        # Show encouragement
        if bot_response.get("encouragement"):
            correction_parts.append(f"💡 **Encouragement:**\n{bot_response['encouragement']}")
        
        # Always show user info if available
        if tutor.user_info.get("name"):
            info_parts = []
            if tutor.user_info.get("name"): info_parts.append(f"Name: {tutor.user_info['name']}")
            if tutor.user_info.get("country"): info_parts.append(f"Country: {tutor.user_info['country']}")
            if tutor.user_info.get("level"): info_parts.append(f"Level: {tutor.user_info['level']}")
            if info_parts:
                correction_parts.append(f"👤 **Your Profile:**\n{' | '.join(info_parts)}")
        
        # If still no corrections, show a default message
        if not correction_parts:
            correction_parts.append("🎯 **Feedback:** Keep practicing! Sam is analyzing your English and will provide feedback soon.")
        
        # Create the new corrections text
        new_correction_text = "\n\n".join(correction_parts)
        timestamp = f"[{user_message[:30]}...]" if len(user_message) > 30 else f"[{user_message}]"
        
        if new_corrections:
            new_corrections = new_corrections + f"\n\n--- Latest Response {timestamp} ---\n{new_correction_text}"
        else:
            new_corrections = f"--- Latest Response {timestamp} ---\n{new_correction_text}"
        
        return history, audio_response, new_transcript, new_corrections
    except Exception as e:
        print(f"Error in process_audio: {str(e)}")
        return history, None, transcript, corrections

def submit_recording(audio, history, transcript, corrections):
    return process_audio(audio, history, transcript, corrections)

def clear_chat():
    global tutor
    tutor = EnglishTutor()
    return initialize_chat()

with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🎓 English Learning Assistant with Sam")
    gr.Markdown("🎤 **Record your voice** - Sam will automatically respond when you finish recording and help improve your English!")
    
    with gr.Row():
        with gr.Column(scale=3):
            chatbot = gr.Chatbot(
                height=500,
                show_label=False,
                type='messages',
                avatar_images=("🎤", "🤖")
            )
            
            with gr.Row():
                with gr.Column(scale=1):
                    audio_input = gr.Audio(
                        label="🎙️ Record your voice (auto-submits when finished)",
                        type="filepath",
                        show_label=True
                    )
                with gr.Column(scale=1):
                    audio_output = gr.Audio(
                        label="🔊 Sam's response",
                        type="filepath",
                        show_label=True,
                        autoplay=True
                    )
        
        with gr.Column(scale=2):
            gr.Markdown("### 📝 Live Transcript")
            transcript_display = gr.Textbox(
                lines=10,
                max_lines=10,
                show_label=False,
                interactive=False,
                placeholder="Your conversation will appear here...",
                container=True
            )
            
            gr.Markdown("### 📚 Learning Corner")
            corrections_display = gr.Textbox(
                lines=8,
                max_lines=8,
                show_label=False,
                interactive=False,
                placeholder="Grammar corrections, vocabulary suggestions, and level assessment will appear here...",
                container=True
            )
    
    with gr.Row():
        clear_btn = gr.Button("🔄 Start New Conversation", variant="secondary", size="lg")
        gr.Markdown("💡 **Tip**: Sam will actively guide the conversation and provide personalized feedback!")
    
    # Auto-submit when audio is recorded
    audio_input.change(
        process_audio,
        inputs=[audio_input, chatbot, transcript_display, corrections_display],
        outputs=[chatbot, audio_output, transcript_display, corrections_display]
    )
    
    clear_btn.click(
        clear_chat,
        outputs=[chatbot, audio_output, transcript_display, corrections_display]
    )
    
    demo.load(
        initialize_chat,
        outputs=[chatbot, audio_output, transcript_display, corrections_display]
    )

if __name__ == "__main__":
    demo.launch()