app.py

import torch
from flask import Flask, render_template, request, jsonify
import os
from transformers import pipeline
from gtts import gTTS
from pydub import AudioSegment
from pydub.silence import detect_nonsilent
from transformers import AutoConfig  # Import AutoConfig for the config object
import time
from waitress import serve
from simple_salesforce import Salesforce
import requests  # Import requests for exception handling

app = Flask(__name__)

# Use whisper-small for faster processing and better speed
device = "cuda" if torch.cuda.is_available() else "cpu"

# Create config object to set timeout and other parameters
config = AutoConfig.from_pretrained("openai/whisper-small")
config.update({"timeout": 60})  # Set timeout to 60 seconds

# Function to generate audio prompts
def generate_audio_prompt(text, filename):
    tts = gTTS(text=text, lang="en")
    tts.save(os.path.join("static", filename))

# Generate required voice prompts
prompts = {
    "welcome": "Welcome to Biryani Hub.",
    "ask_name": "Tell me your name.",
    "ask_email": "Please provide your email address.",
    "thank_you": "Thank you for registration."
}

for key, text in prompts.items():
    generate_audio_prompt(text, f"{key}.mp3")

# Symbol mapping for proper recognition
SYMBOL_MAPPING = {
    "at the rate": "@",
    "at": "@",
    "dot": ".",
    "underscore": "_",
    "hash": "#",
    "plus": "+",
    "dash": "-",
    "comma": ",",
    "space": " "
}

# Function to convert audio to WAV format
def convert_to_wav(input_path, output_path):
    try:
        audio = AudioSegment.from_file(input_path)
        audio = audio.set_frame_rate(16000).set_channels(1)  # Convert to 16kHz, mono
        audio.export(output_path, format="wav")
    except Exception as e:
        print(f"Error: {str(e)}")
        raise Exception(f"Audio conversion failed: {str(e)}")

# Function to check if audio contains actual speech
def is_silent_audio(audio_path):
    audio = AudioSegment.from_wav(audio_path)
    nonsilent_parts = detect_nonsilent(audio, min_silence_len=500, silence_thresh=audio.dBFS-16)  # Reduced silence duration
    print(f"Detected nonsilent parts: {nonsilent_parts}")
    return len(nonsilent_parts) == 0  # If no speech detected

# Salesforce connection details
try:
    print("Attempting to connect to Salesforce...")
    sf = Salesforce(username='[email protected]', password='Sati@1020', security_token='sSSjyhInIsUohKpG8sHzty2q')
    print("Connected to Salesforce successfully!")
    print("User Info:", sf.UserInfo)  # Log the user info to verify the connection
except Exception as e:
    print(f"Failed to connect to Salesforce: {str(e)}")

# Function to create Salesforce record
def create_salesforce_record(name, email, phone_number):
    try:
        # Attempt to create a record in Salesforce with correct field API names
        customer_login = sf.Customer_Login__c.create({
            'Name': name,  # Standard field (name)
            'Email__c': email,  # Custom email field
            'Phone_Number__c': phone_number  # Custom phone number field
        })
        
        # Log the full response from Salesforce
        print(f"Salesforce response: {customer_login}")

        # Check if the response contains an ID (successful creation)
        if customer_login.get('id'):
            print(f"Record created successfully with ID: {customer_login['id']}")
            return customer_login
        else:
            print(f"Record creation failed. Full response: {customer_login}")
            return {"error": f"Record creation failed. Full response: {customer_login}"}
    except Exception as e:
        # Catch and log any exceptions during record creation
        error_message = str(e)
        print(f"Error creating Salesforce record: {error_message}")
        return {"error": f"Failed to create record in Salesforce: {error_message}"}

# Correct the indentation here
@app.route("/submit", methods=["POST"])
def submit():
    data = request.json
    name = data.get('name')
    email = data.get('email')
    phone = data.get('phone')

    if not name or not email or not phone:
        return jsonify({'error': 'Missing data'}), 400

    try:
        # Create Salesforce record
        customer_login = sf.Customer_Login__c.create({
            'Name': name,
            'Email__c': email,
            'Phone_Number__c': phone
        })

        if customer_login.get('id'):
            return jsonify({'success': True})
        else:
            return jsonify({'error': 'Failed to create record'}), 500

    except Exception as e:
        return jsonify({'error': str(e)}), 500

@app.route("/")
def index():
    return render_template("index.html")

@app.route("/transcribe", methods=["POST"])
def transcribe():
    if "audio" not in request.files:
        print("No audio file provided")
        return jsonify({"error": "No audio file provided"}), 400

    audio_file = request.files["audio"]
    input_audio_path = os.path.join("static", "temp_input.wav")
    output_audio_path = os.path.join("static", "temp.wav")
    audio_file.save(input_audio_path)

    try:
        # Convert to WAV
        convert_to_wav(input_audio_path, output_audio_path)

        # Check for silence
        if is_silent_audio(output_audio_path):
            return jsonify({"error": "No speech detected. Please try again."}), 400
        else:
            print("Audio contains speech, proceeding with transcription.")

        # Use Whisper ASR model for transcription
        result = None
        retry_attempts = 3
        for attempt in range(retry_attempts):
            try:
                result = pipeline("automatic-speech-recognition", model="openai/whisper-small", device=0 if torch.cuda.is_available() else -1, config=config)
                print(f"Transcribed text: {result['text']}")
                break
            except requests.exceptions.ReadTimeout:
                print(f"Timeout occurred, retrying attempt {attempt + 1}/{retry_attempts}...")
                time.sleep(5)

        if result is None:
            return jsonify({"error": "Unable to transcribe audio after retries."}), 500

        transcribed_text = result["text"].strip().capitalize()
        print(f"Transcribed text: {transcribed_text}")

        # Extract name, email, and phone number from the transcribed text
        parts = transcribed_text.split()
        name = parts[0] if len(parts) > 0 else "Unknown Name"
        email = parts[1] if '@' in parts[1] else "[email protected]"
        phone_number = parts[2] if len(parts) > 2 else "0000000000"
        print(f"Parsed data - Name: {name}, Email: {email}, Phone Number: {phone_number}")

        # Create record in Salesforce
        salesforce_response = create_salesforce_record(name, email, phone_number)

        # Log the Salesforce response
        print(f"Salesforce record creation response: {salesforce_response}")

        # Check if the response contains an error
        if "error" in salesforce_response:
            print(f"Error creating record in Salesforce: {salesforce_response['error']}")
            return jsonify(salesforce_response), 500

        # If creation was successful, return the details
        return jsonify({"text": transcribed_text, "salesforce_record": salesforce_response})

    except Exception as e:
        print(f"Error in transcribing or processing: {str(e)}")
        return jsonify({"error": f"Speech recognition error: {str(e)}"}), 500

# Start Production Server
if __name__ == "__main__":
    serve(app, host="0.0.0.0", port=7860)