app.py

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_community.graphs import Neo4jGraph
from typing import List, Tuple
from pydantic import BaseModel, Field
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


# Setup Neo4j connection
graph = Neo4jGraph(
    url="neo4j+s://6457770f.databases.neo4j.io",
    username="neo4j",
    password="Z10duoPkKCtENuOukw3eIlvl0xJWKtrVSr-_hGX1LQ4"
)

# Define entity extraction and retrieval functions
class Entities(BaseModel):
    names: List[str] = Field(
        ..., description="All the person, organization, or business entities that appear in the text"
    )

entity_prompt = ChatPromptTemplate.from_messages([
    ("system", "You are extracting organization and person entities from the text."),
    ("human", "Use the given format to extract information from the following input: {question}"),
])

chat_model = ChatOpenAI(temperature=0, model_name="gpt-4o", api_key=os.environ['OPENAI_API_KEY'])
entity_chain = entity_prompt | chat_model.with_structured_output(Entities)

def remove_lucene_chars(input: str) -> str:
    return input.translate(str.maketrans({
        "\\": r"\\", "+": r"\+", "-": r"\-", "&": r"\&", "|": r"\|", "!": r"\!",
        "(": r"\(", ")": r"\)", "{": r"\{", "}": r"\}", "[": r"\[", "]": r"\]",
        "^": r"\^", "~": r"\~", "*": r"\*", "?": r"\?", ":": r"\:", '"': r'\"',
        ";": r"\;", " ": r"\ "
    }))

def generate_full_text_query(input: str) -> str:
    full_text_query = ""
    words = [el for el in remove_lucene_chars(input).split() if el]
    for word in words[:-1]:
        full_text_query += f" {word}~2 AND"
    full_text_query += f" {words[-1]}~2"
    return full_text_query.strip()

# Setup logging to a file to capture debug information
logging.basicConfig(filename='neo4j_retrieval.log', level=logging.DEBUG, format='%(asctime)s - %(levelname)s - %(message)s')

def structured_retriever(question: str) -> str:
    result = ""
    entities = entity_chain.invoke({"question": question})
    for entity in entities.names:
        response = graph.query(
            """CALL db.index.fulltext.queryNodes('entity', $query, {limit:2})
            YIELD node,score
            CALL {
              WITH node
              MATCH (node)-[r:!MENTIONS]->(neighbor)
              RETURN node.id + ' - ' + type(r) + ' -> ' + neighbor.id AS output
              UNION ALL
              WITH node
              MATCH (node)<-[r:!MENTIONS]-(neighbor)
              RETURN neighbor.id + ' - ' + type(r) + ' -> ' +  node.id AS output
            }
            RETURN output LIMIT 50
            """,
            {"query": generate_full_text_query(entity)},
        )
        result += "\n".join([el['output'] for el in response])
    return result

def retriever_neo4j(question: str):
    structured_data = structured_retriever(question)
    logging.debug(f"Structured data: {structured_data}")
    return structured_data

# Setup for condensing the follow-up questions
_template = """Given the following conversation and a follow-up question, rephrase the follow-up question to be a standalone question,
in its original language.
Chat History:
{chat_history}
Follow Up Input: {question}
Standalone question:"""

CONDENSE_QUESTION_PROMPT = PromptTemplate.from_template(_template)

def _format_chat_history(chat_history: list[tuple[str, str]]) -> list:
    buffer = []
    for human, ai in chat_history:
        buffer.append(HumanMessage(content=human))
        buffer.append(AIMessage(content=ai))
    return buffer

_search_query = RunnableBranch(
    (
        RunnableLambda(lambda x: bool(x.get("chat_history"))).with_config(
            run_name="HasChatHistoryCheck"
        ),
        RunnablePassthrough.assign(
            chat_history=lambda x: _format_chat_history(x["chat_history"])
        )
        | CONDENSE_QUESTION_PROMPT
        | ChatOpenAI(temperature=0, api_key=os.environ['OPENAI_API_KEY'])
        | StrOutputParser(),
    ),
    RunnableLambda(lambda x: x["question"]),
)


template = """I am a guide for Birmingham, Alabama. I can provide recommendations and insights about the city, including events and activities.
Ask your question directly, and I'll provide a precise and quick,short and crisp response in a conversational way without any Greet.
{context}
Question: {question}
Answer:"""


qa_prompt = ChatPromptTemplate.from_template(template)

# Define the chain for Neo4j-based retrieval and response generation
chain_neo4j = (
    RunnableParallel(
        {
            "context": RunnableLambda(lambda x: retriever_neo4j(x["question"])),
            "question": RunnablePassthrough(),
        }
    )
    | ChatPromptTemplate.from_template("Answer: {context} Question: {question}")
    | chat_model
    | StrOutputParser()
)

# Define the function to query Neo4j and get a response
def get_response(question):
    try:
        return chain_neo4j.invoke({"question": question})
    except Exception as e:
        return f"Error: {str(e)}"



# 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
        return audio_path
    else:
        return None

# Define ASR model for speech-to-text
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
)

# Set up logging to a file to capture debug information
logging.basicConfig(filename='neo4j_retrieval.log', level=logging.DEBUG, format='%(asctime)s - %(levelname)s - %(message)s')

# Function to handle voice input, generate response from Neo4j, and return audio output
def handle_voice_to_voice(audio):
    try:
        # Transcribe audio input to text
        sr, y = audio
        y = y.astype(np.float32)
        y = y / np.max(np.abs(y))  # Normalize audio to range [-1.0, 1.0]

        logging.debug(f"Audio sample rate: {sr}")
        logging.debug(f"Audio data: {y[:100]}")  # Log first 100 samples for brevity

        # Process the audio data with Whisper ASR
        result = pipe_asr({"array": y, "sampling_rate": sr}, return_timestamps=False)
        question = result.get("text", "")

        logging.debug(f"Transcribed question: {question}")

        if not question:
            return "No transcription available, please try again.", ""

        # Get response using the transcribed question
        response = get_response(question)
        logging.debug(f"Response from Neo4j and GPT: {response}")

        # Generate audio from the response
        audio_path = generate_audio_elevenlabs(response)
        logging.debug(f"Generated audio path: {audio_path}")

        # Return the transcription and the audio path
        return audio_path, question
    except Exception as e:
        logging.error(f"Error in handle_voice_to_voice: {e}")
        return "Error processing the audio, please try again.", ""


# Define the Gradio interface
with gr.Blocks(theme="rawrsor1/Everforest") as demo:
    audio_input = gr.Audio(sources=["microphone"], type='numpy', streaming=True, label="Speak to Ask")
    submit_voice_btn = gr.Button("Submit Voice")
    transcription_textbox = gr.Textbox(label="Transcription", interactive=False)
    audio_output = gr.Audio(label="Response Audio", type="filepath", autoplay=True, interactive=False)

    # Interactions for Submit Voice Button
    submit_voice_btn.click(
        fn=handle_voice_to_voice,
        inputs=audio_input,
        outputs=[audio_output, transcription_textbox]
    ).then(
        fn=lambda: "",
        outputs=transcription_textbox
    )

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