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 waitress import serve
from simple_salesforce import Salesforce

app = Flask(__name__)

# Use whisper-small for faster processing and better speed
device = "cuda" if torch.cuda.is_available() else "cpu"
asr_model = pipeline("automatic-speech-recognition", model="openai/whisper-small", device=0 if device == "cuda" else -1)

# 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:
        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
    return len(nonsilent_parts) == 0  # If no speech detected

# Salesforce connection details
sf = Salesforce(username='[email protected]', password='Sati@1020', security_token='sSSjyhInIsUohKpG8sHzty2q')

# Debugging - Confirming Salesforce connection
print("Connected to Salesforce, checking user info...")
print(sf.UserInfo)  # This will give you the user info if the connection is successful

# Function to create Salesforce record
def create_salesforce_record(name, email, phone_number):
    try:
        # Attempt to create a record in Salesforce
        customer_login = sf.Customer_Login__c.create({
            'Name': name,
            'Email__c': email,
            'Phone_Number__c': phone_number
        })
        
        # Log the response from Salesforce
        print(f"Salesforce response: {customer_login}")
        
        if customer_login.get('id'):
            print(f"Record created successfully with ID: {customer_login['id']}")
        else:
            print("No ID returned, record creation may have failed.")
        
        return 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}"}



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

@app.route("/transcribe", methods=["POST"])
def transcribe():
    if "audio" not in request.files:
        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

        # Use Whisper ASR model for transcription
        result = asr_model(output_audio_path, generate_kwargs={"language": "en"})
        transcribed_text = result["text"].strip().capitalize()

        # Now, let's split the transcribed text into name, email, and phone number (basic example)
        parts = transcribed_text.split()  # This is a simplistic approach; you may need a better parsing mechanism
        name = parts[0]  # Assuming first word is the name
        email = parts[1] if '@' in parts[1] else "[email protected]"  # Very basic email extraction
        phone_number = parts[2] if len(parts) > 2 else "0000000000"  # Basic phone number assumption

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

        # 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
        
        print(f"Salesforce Response: {salesforce_response}")
        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)