import gradio as gr
import os
import logging
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI
from langchain_core.messages import AIMessage, HumanMessage
from langchain_core.runnables import (
    RunnableBranch,
    RunnableLambda,
    RunnablePassthrough,
    RunnableParallel,
)
from langchain_core.prompts.prompt import PromptTemplate
import requests
import tempfile
from langchain.memory import ConversationBufferWindowMemory
import time
import logging
from langchain.chains import ConversationChain
import torch
import torchaudio
from transformers import pipeline, AutoModelForSpeechSeq2Seq, AutoProcessor
import numpy as np
import threading
from langchain_openai import OpenAIEmbeddings
from langchain_pinecone import PineconeVectorStore
from langchain.chains import RetrievalQA

embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
def initialize_gpt_model():
    return ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model='gpt-4o')

gpt_model = initialize_gpt_model()


gpt_embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
gpt_vectorstore = PineconeVectorStore(index_name="radardata10312024", embedding=gpt_embeddings)
gpt_retriever = gpt_vectorstore.as_retriever(search_kwargs={'k': 5})



# Pinecone setup
from pinecone import Pinecone
pc = Pinecone(api_key=os.environ['PINECONE_API_KEY'])


index_name ="radardata10312024"
vectorstore = PineconeVectorStore(index_name=index_name, embedding=embeddings)
retriever = vectorstore.as_retriever(search_kwargs={'k': 5})

chat_model = ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model='gpt-4o')

#code for history
conversational_memory = ConversationBufferWindowMemory(
        memory_key='chat_history',
        k=10,
        return_messages=True
    )

template =f"""Hello there! As your friendly and knowledgeable guide here in Birmingham, Alabama.Give the short ,precise,crisp and straight-foreward response of maximum 2 sentences and dont greet. 
{{context}}
Question: {{question}}
Helpful Answer:"""

QA_CHAIN_PROMPT= PromptTemplate(input_variables=["context", "question"], template=template)

def build_qa_chain(prompt_template):
    qa_chain = RetrievalQA.from_chain_type(
        llm=chat_model,
        chain_type="stuff",
        retriever=retriever,
        chain_type_kwargs={"prompt": prompt_template}
    )
    return qa_chain  # Return the qa_chain object

# Instantiate the QA Chain using the defined prompt template
qa_chain = build_qa_chain(QA_CHAIN_PROMPT)

# Define the function to clear input and output
def clear_fields():
    return [],"",None

# Function to generate audio with Eleven Labs TTS
def generate_audio_elevenlabs(text):
    XI_API_KEY = os.environ['ELEVENLABS_API']
    VOICE_ID = 'ehbJzYLQFpwbJmGkqbnW'
    tts_url = f"https://api.elevenlabs.io/v1/text-to-speech/{VOICE_ID}/stream"
    headers = {
        "Accept": "application/json",
        "xi-api-key": XI_API_KEY
    }
    data = {
        "text": str(text),
        "model_id": "eleven_multilingual_v2",
        "voice_settings": {
            "stability": 1.0,
            "similarity_boost": 0.0,
            "style": 0.60,
            "use_speaker_boost": False
        }
    }
    response = requests.post(tts_url, headers=headers, json=data, stream=True)
    if response.ok:
        with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as f:
            for chunk in response.iter_content(chunk_size=1024):
                if chunk:
                    f.write(chunk)
            audio_path = f.name
        logging.debug(f"Audio saved to {audio_path}")
        return audio_path  # Return audio path for automatic playback
    else:
        logging.error(f"Error generating audio: {response.text}")
        return None



import time

# Main function to handle mode selection with character-by-character streaming
def handle_mode_selection(mode, chat_history, question):
    if mode == "Normal Chatbot":
        chat_history.append((question, ""))  # Append user question with an empty response initially

        # Get response from Pinecone using the qa_chain
        response = qa_chain({"query": question, "context": ""})
        response_text = response['result']

        # Stream each character in the response text to the chat history
        for i, char in enumerate(response_text):
            chat_history[-1] = (question, chat_history[-1][1] + char)  # Update the last message
            yield chat_history, "", None  # Yield updated chat history
            time.sleep(0.05)  # Small delay to simulate streaming

    elif mode == "Voice to Voice Conversation":
        response_text = qa_chain({"query": question, "context": ""})['result']
        audio_path = generate_audio_elevenlabs(response_text)
        yield [], "", audio_path  # Only output the audio response without updating chatbot history




# Function to add a user's message to the chat history and clear the input box
def add_message(history, message):
    if message.strip():
        history.append((message, ""))  # Add the user's message to the chat history only if it's not empty
    return history, ""  # Clear the input box
   
# Define function to generate a streaming response
def chat_with_bot(messages):
    user_message = messages[-1][0]  # Get the last user message (input)
    messages[-1] = (user_message, "")  # Prepare a placeholder for the bot's response
    
    response = get_response(user_message)  # Assume `get_response` is a generator function
    
    # Stream each character in the response and update the history progressively
    for character in response:
        messages[-1] = (user_message, messages[-1][1] + character)
        yield messages  # Stream each updated chunk
        time.sleep(0.05)  # Adjust delay as needed for real-time effect

    yield messages  # Final yield to complete the response


    
# Function to generate audio with Eleven Labs TTS from the last bot response
def generate_audio_from_last_response(history):
    # Get the most recent bot response from the chat history
    if history and len(history) > 0:
        recent_response = history[-1][1]  # The second item in the tuple is the bot response text
        if recent_response:
            return generate_audio_elevenlabs(recent_response)
    return None

# Define example prompts
examples = [
    ["what are the tree care services at alabama?"],
    ["where from i studies undergrade in marketing from alabama?"],
    ["what from i get tourism recreation center?"],
    ["where from i will get a retail loan and from which institute?"],
    ["where i will look for good dentist at alabama?"]
    
]

# Function to insert the prompt into the textbox when clicked
def insert_prompt(current_text, prompt):
    return prompt[0] if prompt else current_text


# Define the ASR model with Whisper
model_id = 'openai/whisper-large-v3'
device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype).to(device)
processor = AutoProcessor.from_pretrained(model_id)

pipe_asr = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    max_new_tokens=128,
    chunk_length_s=15,
    batch_size=16,
    torch_dtype=torch_dtype,
    device=device,
    return_timestamps=True
)

# Define the function to reset the state after 10 seconds
def auto_reset_state():
    time.sleep(5)
    return None, ""  # Reset the state and clear input text


def transcribe_function(stream, new_chunk):
    try:
        sr, y = new_chunk[0], new_chunk[1]
    except TypeError:
        print(f"Error chunk structure: {type(new_chunk)}, content: {new_chunk}")
        return stream, "", None

    # Ensure y is not empty and is at least 1-dimensional
    if y is None or len(y) == 0:
        return stream, "", None

    y = y.astype(np.float32)
    max_abs_y = np.max(np.abs(y))
    if max_abs_y > 0:
        y = y / max_abs_y

    # Ensure stream is also at least 1-dimensional before concatenation
    if stream is not None and len(stream) > 0:
        stream = np.concatenate([stream, y])
    else:
        stream = y

    # Process the audio data for transcription
    result = pipe_asr({"array": stream, "sampling_rate": sr}, return_timestamps=False)
    full_text = result.get("text", "")

    # Start a thread to reset the state after 10 seconds
    threading.Thread(target=auto_reset_state).start()

    return stream, full_text, full_text



# Define the function to clear the state and input text
def clear_transcription_state():
    return None, ""



with gr.Blocks(theme="rawrsor1/Everforest") as demo:
    chatbot = gr.Chatbot([], elem_id="RADAR", bubble_full_width=False)
    with gr.Row():
        with gr.Column():
            mode_selection = gr.Radio(
                choices=["Normal Chatbot", "Voice to Voice Conversation"],
                label="Mode Selection",
                value="Normal Chatbot"
            )
    with gr.Row():
        with gr.Column():
            question_input = gr.Textbox(label="Ask a Question", placeholder="Type your question here...")
            audio_input = gr.Audio(sources=["microphone"], streaming=True, type='numpy', every=0.1, label="Speak to Ask")
            submit_voice_btn = gr.Button("Submit Voice")

        with gr.Column():
            audio_output = gr.Audio(label="Audio", type="filepath", autoplay=True, interactive=False)
    
    with gr.Row():
        with gr.Column():
            get_response_btn = gr.Button("Get Response")
        with gr.Column():
            clear_state_btn = gr.Button("Clear State")
        with gr.Column():
            generate_audio_btn = gr.Button("Generate Audio")
        with gr.Column():
            clean_btn = gr.Button("Clean")

    with gr.Row():
        with gr.Column():
            gr.Markdown("<h1 style='color: red;'>Example Prompts</h1>", elem_id="Example-Prompts")
            gr.Examples(examples=examples, fn=insert_prompt, inputs=question_input, outputs=question_input, api_name="api_insert_example")

  
    # Define interactions for the Get Response button
    get_response_btn.click(
        fn=handle_mode_selection,
        inputs=[mode_selection, chatbot, question_input],
        outputs=[chatbot, question_input, audio_output],
        api_name="api_add_message_on_button_click"
        )

    
    

    question_input.submit(
        fn=handle_mode_selection,
        inputs=[mode_selection, chatbot, question_input],
        outputs=[chatbot, question_input, audio_output],
        api_name="api_add_message_on_enter"
    )

    
    submit_voice_btn.click(
        fn=handle_mode_selection,
        inputs=[mode_selection, chatbot, question_input],
        outputs=[chatbot, question_input, audio_output],
        api_name="api_voice_to_voice_translation"
    )


    
    # Speech-to-Text functionality
    state = gr.State()
    audio_input.stream(
        transcribe_function,
        inputs=[state, audio_input],
        outputs=[state, question_input],
        api_name="api_voice_to_text"
    )    
    
    generate_audio_btn.click(
        fn=generate_audio_from_last_response,
        inputs=chatbot,
        outputs=audio_output,
        api_name="api_generate_text_to_audio"
    )
    
    clean_btn.click(
        fn=clear_fields,
        inputs=[],
        outputs=[chatbot, question_input, audio_output],
        api_name="api_clear_textbox"
    )

    # Clear state interaction
    clear_state_btn.click(
        fn=clear_transcription_state,
        outputs=[question_input, state],
        api_name="api_clean_state_transcription"
    )

# Launch the Gradio interface
demo.launch(show_error=True,share=True)