app.py

import torch
from flask import Flask, render_template, request, jsonify
import os
import re
from transformers import pipeline
from gtts import gTTS
from pydub import AudioSegment
from pydub.silence import detect_nonsilent
from waitress import serve

app = Flask(__name__)

# ✅ Load Whisper ASR Model Correctly
device = "cuda" if torch.cuda.is_available() else "cpu"
asr_model = pipeline("automatic-speech-recognition", model="openai/whisper-large-v3", device=0 if device == "cuda" else -1)

# Function to generate voice 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")

# ✅ Ensure Proper Audio 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)}")

# ✅ Check for Silence
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)
    return len(nonsilent_parts) == 0

@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 audio to proper format
        convert_to_wav(input_audio_path, output_audio_path)

        # ✅ Check for silent audio
        if is_silent_audio(output_audio_path):
            return jsonify({"error": "No speech detected. Please try again."}), 400

        # ✅ Transcribe Using Whisper ASR
        result = asr_model(output_audio_path, generate_kwargs={"language": "en"})
        transcribed_text = result["text"].strip().capitalize()

        return jsonify({"text": transcribed_text})
    except Exception as 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)