app.py

import os
import speech_recognition as sr
import fitz  # PyMuPDF
from transformers import AutoTokenizer, AutoModel
import torch
import faiss
import numpy as np
from gtts import gTTS
from pydub import AudioSegment
from groq import Groq
from dotenv import load_dotenv
import gradio as gr

# Load environment variables
load_dotenv()

# Initialize Groq API client
client = Groq(api_key=os.getenv("GROQ_API_KEY"))

# Initialize model and tokenizer for embedding
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
model = AutoModel.from_pretrained("bert-base-uncased")

# Initialize vector database
dimension = 768  # Size of BERT embeddings
index = faiss.IndexFlatL2(dimension)

# Folder path containing PDFs
pdf_folder_path = "agriculture_pdfs"

# Function to convert audio file to text
def audio_to_text(audio_file_path):
    recognizer = sr.Recognizer()
    try:
        with sr.AudioFile(audio_file_path) as source:
            audio = recognizer.record(source)
        text = recognizer.recognize_google(audio)
        return text
    except sr.UnknownValueError:
        return None
    except sr.RequestError:
        return None

# Function to convert audio to WAV format
def convert_to_wav(audio_file_path):
    if not audio_file_path:
        raise ValueError("Invalid audio file path")
    try:
        audio = AudioSegment.from_file(audio_file_path)
        wav_path = "temp_audio.wav"
        audio.export(wav_path, format="wav")
        return wav_path
    except Exception:
        return None

# Function to extract text from a PDF file
def extract_text_from_pdf(pdf_file):
    text = ""
    try:
        pdf_document = fitz.open(pdf_file)
        for page_num in range(len(pdf_document)):
            page = pdf_document.load_page(page_num)
            text += page.get_text()
    except Exception:
        pass
    return text

# Function to embed text using a transformer model
def embed_text(texts, model, tokenizer):
    try:
        inputs = tokenizer(texts, return_tensors='pt', truncation=True, padding=True)
        with torch.no_grad():
            embeddings = model(**inputs).last_hidden_state.mean(dim=1).numpy()
        return embeddings
    except Exception:
        return np.array([])  # Return empty array on error

# Function to convert text to speech
def text_to_speech(text, output_file):
    try:
        tts = gTTS(text=text, lang='en')
        tts.save(output_file)
        return output_file
    except Exception:
        return None

# Read all PDF files from the specified folder
pdf_paths = [os.path.join(pdf_folder_path, f) for f in os.listdir(pdf_folder_path) if f.endswith('.pdf')]

texts = []
for path in pdf_paths:
    pdf_text = extract_text_from_pdf(path)
    if pdf_text:
        texts.append(pdf_text)

# Embed PDF texts and add to vector database
embeddings = embed_text(texts, model, tokenizer)
if embeddings.size > 0:
    index.add(embeddings)

def process_audio(audio_file):
    if audio_file is None:
        return "No audio file provided", None

    audio_file_path = audio_file if isinstance(audio_file, str) else audio_file.name
    wav_path = convert_to_wav(audio_file_path)
    if wav_path is None:
        return "Error converting audio file to WAV format", None

    text = audio_to_text(wav_path)
    if not text:
        return "No valid text extracted from audio", None

    try:
        audio_embedding = embed_text([text], model, tokenizer)[0]
        if audio_embedding.size == 0:
            return "Error generating embedding for the audio text", None
        
        distances, indices = index.search(np.array([audio_embedding]), k=5)
        relevant_texts = [texts[idx] for idx in indices[0]]
        combined_text = " ".join(relevant_texts)
        if len(combined_text) > 1000:
            combined_text = combined_text[:1000]

        if not combined_text.strip():
            return "No relevant information found in the PDFs", None
        
        prompt = (
            f"The user has asked a query related to agricultural practices: {text}. "
            f"Here are relevant excerpts from the Better Crops South Asia document: {combined_text}. "
            "Based on this information, please provide accurate advice related to sustainable crop management, pest control, irrigation practices, and any recommendations for improving crop yield in the South Asian region."
        )

        chat_completion = client.chat.completions.create(
            messages=[
                {
                    "role": "user",
                    "content": prompt,
                }
            ],
            model="llama-3.1-70b-versatile",
        )
        response = chat_completion.choices[0].message.content
        output_file = "advice.mp3"
        output_path = text_to_speech(response, output_file)

        if output_path is None:
            return "Error generating speech output", None

        return response, output_path
    except Exception:
        return "An error occurred while processing the audio", None

# Enhanced Gradio interface customization
iface = gr.Interface(
    fn=process_audio,
    inputs=gr.Audio(type="filepath"),
    outputs=[gr.Textbox(label="Advice", lines=10), gr.Audio(label="Advice Audio")],
    title="🌾 BetterCrops: Agriculture Support for Farmers",
    description=(
        "💡 **BetterCrops** is designed to assist farmers with their crops by analyzing agricultural PDFs "
        "and generating personalized audio advice based on your voice queries."
    ),
    article=(
        "<div style='text-align: center; color: #003f6e;'>"
        "<h1 style='font-size: 36px; font-weight: bold;'>BetterCrops</h1>"
        "<h3 style='font-size: 24px; font-weight: normal;'>Empowering Farmers with AI-driven Insights</h3>"
        "</div>"
    ),
    theme="grass",
    css=(
        """
        body {
            background-color: #f0f5e9;
            color: #2f4f2f;
            font-family: 'Helvetica Neue', sans-serif;
        }
        h1, h3 {
            color: #003f6e;
        }
        .gradio-container {
            padding: 20px;
            background: linear-gradient(135deg, #a3cfba 0%, #e8f5e9 100%);
            border-radius: 15px;
        }
        .gradio-inputs, .gradio-outputs {
            margin: 20px;
            padding: 20px;
            background-color: #ffffff;
            border-radius: 10px;
            box-shadow: 0px 4px 8px rgba(0, 0, 0, 0.1);
        }
        """
    )
)

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