app.py

import gradio as gr
import whisper
import os
import tempfile
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
from langchain_community.llms import HuggingFaceEndpoint

# Load Whisper model (change to "base", "small", or "medium" for better accuracy)
model = whisper.load_model("base")

# Model config for Hugging Face Inference API
hub = {
    "HF_MODEL_ID": "mistralai/Mistral-7B-Instruct-v0.3",  
    "HF_TASK": "text-generation",
    "HF_API_TOKEN": os.environ.get("HUGGINGFACE_API_TOKEN")
}

# Global state
vector_db = None
qa_chain = None
chat_memory = None

# Transcribe and set up RAG
def transcribe_and_setup(audio_file_path):
    global vector_db, qa_chain, chat_memory

    if audio_file_path is None:
        return "No audio uploaded.", []

    result = model.transcribe(audio_file_path)
    transcript = result["text"]

    # Split and embed
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=64)
    splits = text_splitter.create_documents([transcript])
    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
    vector_db = FAISS.from_documents(splits, embeddings)

    # QA setup
    chat_memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
    retriever = vector_db.as_retriever()
    llm = HuggingFaceEndpoint(
        repo_id=hub["HF_MODEL_ID"],
        task=hub["HF_TASK"],
        huggingfacehub_api_token=hub["HF_API_TOKEN"],
        temperature=0.5,
        max_new_tokens=512
    )
    qa_chain = ConversationalRetrievalChain.from_llm(llm=llm, retriever=retriever, memory=chat_memory)

    return "Transcription complete! Ready for questions.", []  # Empty chat reset

# Handle conversation
def answer_question(question, chat_history):
    global qa_chain, chat_memory
    if qa_chain is None:
        return "Please upload and process an audio file first.", chat_history

    response = qa_chain.invoke({
        "question": question,
        "chat_history": chat_memory.load_memory_variables({})["chat_history"]
    })

    # Just show back-and-forth messages
    chat_history.append([question, response["answer"]])
    return "", chat_history

# Custom CSS
custom_css = """
.chatbox .message.user, .chatbox .message.bot {
    background-color: #1e3d2f !important;
    color: #ffffff !important;
    border-radius: 10px !important;
    padding: 10px !important;
    margin: 5px !important;
}
.chatbox .message.user::before, .chatbox .message.bot::before {
    content: none !important;
}
"""

# Gradio app
with gr.Blocks(theme=gr.themes.Soft(), css=custom_css) as demo:
    gr.Markdown("## 🎙️ **Conversational Audio Chatbot**")
    gr.Markdown("Upload an audio file, let the AI process it, and ask any questions!")

    with gr.Row():
        with gr.Column(scale=1):
            audio_input = gr.Audio(type="filepath", label="🎧 Upload Audio")
            transcribe_button = gr.Button("🚀 Process Audio")
            status_output = gr.Textbox(label="🛠️ Status", interactive=False)
        with gr.Column(scale=2):
            chatbot = gr.Chatbot(label="💬 Chat with your audio", elem_classes=["chatbox"])
            question_input = gr.Textbox(label="Type your question", placeholder="Ask about the audio...")
            ask_button = gr.Button("💡 Ask")

    transcribe_button.click(
        fn=transcribe_and_setup,
        inputs=audio_input,
        outputs=[status_output, chatbot]
    )

    ask_button.click(
        fn=answer_question,
        inputs=[question_input, chatbot],
        outputs=[question_input, chatbot]
    )

demo.launch()